Air quality forecasting system for Southeastern Brazil

Introduction

In southeastern Brazil, the Weather Research and Forecasting model with Chemistry (WRF/Chem) system has been used in the analysis of regulated pollutants and air quality forecasting, validated against ground-based measurements and ozone soundings, since 2011. Although the meteorological fields have been well-represented in the modeling, the pollutant concentration data present biases that are probably related to the estimated emissions inventory. Andrade et al. (2004) have shown the impact of considering the official emission inventory for NO_x and VOC in the ozone simulation. The ratio between the two pollutants resulted in no ozone formation. Since 2004, the Brazilian Regional Atmospheric Modeling System with Simplified Photochemical Module (BRAMS-SPM) modeling system has been applied in the Metropolitan Area of São Paulo (MASP) to evaluate the impact that increases in the size and population of urban areas have on the weather-based forecast of air quality.

Studies concerning air quality and its effect on climate and health have been conducted in the MASP since the 1970s, the first such studies focusing on regulated pollutants, concentrations of particles, and speciation. The MASP presents particular characteristics in terms of atmospheric pollutant sources, climate, land use, and land occupation, when compared with other megacities in the world.

In the 1960s, the MASP received large numbers of immigrants from other states in Brazil, attracted by job opportunities related to industrial expansion in the region. As a consequence, there was a degradation of air quality conditions with high concentrations of pollutants, mainly SO₂ and CO, emitted by industrial activities, and there were at that time no controls on the emissions of air pollutants. In the late 1990s, the vehicle fleet showed rapid growth due to tax cuts, the availability of credit, an increase in the size of the working population, and higher incomes. Vehicle emissions are currently the main source of air pollution in the urban areas of Brazil. The light-duty fleet runs on ethanol, gasohol (85% gasoline and 25% hydrous ethanol), or natural gas, whereas heavy-duty vehicles (busses and trucks) run on diesel. There are also the so-called “flex-fuel” vehicles that can run on gasohol or ethanol.

In comparison with other urban areas of Brazil, the MASP has the best network of air quality monitoring stations, which therefore provide data with the highest spatial resolution. Air quality monitoring began in the late 1970s with the measurement of particulate matter of less than 10 μm in diameter (PM₁₀) and SO₂ by the São Paulo State Companhia de Tecnologia de Saneamento Ambiental (CETESB, Environmental Protection Agency). Orsini et al. (1986) were the first to analyze the composition of particulate matter in the MASP, using particle-induced X-ray emission analysis of polycarbonate membranes sampled with cascade impactors and sequential samplers. Sources were identified by receptor modeling, vehicle emissions being identified as the main source of fine particles, which remains the case. The receptor modeling approach has been applied for many years, producing significant results related to source apportionment in the MASP. In other regions of Brazil, air quality issues have only recently come to be considered a major concern. The World Health Organization (WHO) recently evaluated the worldwide impact of particulate matter on health and estimated that, in 2011, 7 million people died as a result of high atmospheric concentrations of particles, mainly in developing countries such as China and India, which are the most populous countries and have high concentrations of pollutants, 4.3 million of those deaths being associated with indoor pollution and 3.7 million being associated with atmospheric pollution from rural and urban sources (WHO, 2014).

The application of air quality models is considered to be a powerful tool for the study, description and forecasting of air pollution in areas affected by urban pollution, biomass burning, and resuspended soil dust, as described by Krzyzanowski et al. (2014). According to those authors, the concentrations of pollutants can be estimated through the use of three-dimensional models, combined with direct measurements and analysis of satellite images. The three-dimensional air quality models describe the chemistry, emissions, and transport of pollutants. The representation of emissions on a grid of adequate size provides a great deal of certainty. The modeling approach taken in different cities in the southeastern region of Brazil has been evolving, and most of the efforts have been focused on the construction of a comprehensive emissions inventory. In the MASP, the emissions inventory is based on vehicle emission factors (dynamometer measurements) and tunnel measurements performed in the city of São Paulo in 2001 and 2004 (Martins et al., 2006; Sánchez-Ccoyllo et al., 2009), as well as on experiments conducted in 2011 (Nogueira et al., 2014; Pérez-martínez et al., 2014).

Due to a lack of information from CETESB and from the Brazilian National Department of Transplantation (DENATRAN), various strategies have been employed in order to determine the spatial distribution of emission sources in the MASP. One such strategy is the use of nocturnal satellite data to describe the distribution of sources by their density, on the basis of pixel density, as described in Martins et al. (2012). After the changes in fuel composition that occurred between 2004 and 2011, that approach was used in regional studies aimed at analyzing the impact that those changes have had on the behavior of ozone concentrations (Vara Vela, 2013). Mazzoli (2013) analyzed how tropospheric ozone concentrations in the MASP would respond to the climate change scenarios described in the Fourth Assessment Report by the Intergovernmental Panel on Climate Change, as well as to projected changes in the local emissions of pollutants, showing that, in 2030 and 2050, changes in the size of the vehicle fleet and in vehicle emission factors would have an effect of greater magnitude than that related to the predicted changes in temperature and humidity.

Although many photochemical modeling studies have been performed in Brazil, most have been conducted in the MASP, although some have been conducted in the city of Rio de Janeiro. Godoy et al. (2009), using receptor modeling, found that vehicle emissions constitute the main source of pollutants in the city of Rio de Janeiro. Luna et al. (2014), using neural network analysis, developed a system for ozone prediction. In all such initiatives, one important objective is to provide a tool that can be used by governmental agencies to predict events of high pollutant concentrations, which will make it possible to issue alerts and prepare emergency rooms to receive cases related to the poor air quality. Other cities in South America, such as Santiago, Chile, have action plans for such eventualities (Saide et al., 2011). In Brazil, there are national air quality standards for regulated pollutants. However, the state of São Paulo has changed its standards to be more restrictive, in accordance with the WHO guidelines (CETESB, 2014). The state standards will be changed gradually, in three phases, the last of which involves setting the maximum allowable concentrations of regulated pollutants to the values recommended by the WHO.

Air quality modeling can be an effective tool to meet the challenge of describing the chemistry of secondary pollutant formation, specifically that of ozone and fine particles. In the MASP, the study of the formation of ozone and other gaseous photochemical compounds started with the off-line use of the California Institute of Technology air quality model (McRae et al., 1982). In two studies employing that model (Andrade et al., 2004; Vivanco and Andrade, 2006), the atmosphere of the MASP was found to have a high potential to form ozone and that the concentrations of nitrogen oxides (NO_x) in the official emissions inventory are twice as high as those that would be expected on the basis of modeling results and the concentrations measured at air quality monitoring stations. Important aspects of ozone formation in the MASP were studied by Martins et al. (2006), who identified the main precursors of tropospheric ozone, and by Martins and Andrade (2008a), who demonstrated an increase in the reactivity of the ozone precursors in the region.

The impact of meteorological conditions on ozone formation in the MASP was evaluated by Sánchez-Ccoyllo et al. (2006, 2007), who used the California Institute of Technology air quality model under NO_x- or volatile organic compound (VOC)-limited atmospheric conditions. Martins and Andrade (2008b) analyzed various scenarios of the use of pure ethanol, gasohol, and diesel in the MASP, concluding that an increase in the use of ethanol, with a consequent decrease in the use of the gasohol currently available in Brazil, would reduce ozone formation. However, when the analysis involved “better” gasohol (with lower fractions of olefins, aromatics, and benzene) the difference between two scenarios (ethanol vs. standard gasohol and ethanol vs. better gasohol) was small and shifted between positive and negative values, indicating that the use of gasohol containing fewer olefins and aromatics would be more efficient in decreasing ozone concentrations than would be a shift toward greater ethanol use. The modeling capability has been increased by experimental studies aimed at improving the emissions inventory and describing the atmospheric processes. Such experiments consist of measurements of ozone precursors (Nogueira et al., 2014); fine and ultrafine particles (Almeida et al., 2014); and emission factors (Nogueira et al., 2014; Pérez-martínez et al., 2014). Combining all the efforts that have been applied to the representation of pollutants with air quality models in the MASP, an operation was implemented in order to forecast the concentrations of ozone, carbon monoxide, and particles. The forecasts are run daily and are available for consultation at the site: http://www.lapat.iag.usp.br/aerossol/wrf9/index.php. The air quality modeling platform established for southeastern Brazil represents efforts to meet a variety of challenges in the atmospheric sciences: the description of the chemistry of reactive species; the characterization of transport by advection and diffusion; and the development of reliable emissions inventories. Air quality is being forecast with two different tools: the WRF/Chem (Grell et al., 2005); and the BRAMS-SPM (Freitas et al., 2005, 2009). The modeling is an example of combination of measurements of emission factors and ambient concentrations, and the spatial representation of emission using the traffic and street maps as a proxy, to construct the module of mobile emission. This modeling approach can be considered for other places impacted by the pollutants emitted by the vehicular fleet that do not have the spatial and temporal distribution of the mobile sources.

Materials and Methods

The approach adopted to provide the air quality forecasting for the Southeast region of Brazil was the use of eulerian photochemical modeling. Two modeling systems are being used: WRF-Chem [Weather Research and Forecasting System with Chemistry, described in Grell et al. (2005)] and BRAMS-SPM [Brazilian developments on the Regional Atmospheric Modeling System with Simplified Photochemical Module, described by Freitas et al. (2005) and Freitas et al. (2009)]. The implementation of the modeling system had the objective of forecasting regulated pollutants and the study of secondary pollutants formation and health impacts. The photochemical modeling system was integrated to an emission module developed to describe the real conditions of traffic distribution over the modeled area.

Study Region

The modeling area includes the states of São Paulo and Rio de Janeiro, as well as parts of the states of Paraná, Minas Gerais, Espirito Santo e Santa Catarina (Figure 1). It is the most industrialized area in the country, comprising the metropolitan areas of São Paulo (19,683,975 inhabitants), Rio de Janeiro (11,835,708 inhabitants), Belo Horizonte (5,414,701 inhabitants), and other major cities. The region accounts for 55% of the national gross domestic product, being the state of São Paulo alone accounting for approximately 33% of it. It encompasses 150 points in longitude and 100 points in latitude. The model has a horizontal grid spacing of approximately 9 km, 35 vertical levels, and 84 h of integration. The horizontal grid spacing was established based on the size of the urbanized area (approximately 80 km in the W–E direction and 40 km in the N–S direction) in order to satisfactorily represent mesoscale circulations due to the urban heat island, the local topography and the sea-breeze that are frequently observed in the region and are fundamental to the processes involving pollution dispersion, such as horizontal advection, re-circulation, vertical transport, deposition, and others.

Figure 1. Area considered in the simulations, showing elevations (m).

BRAMS-SPM Modeling

As part of Brazilian efforts to develop numerical tools for weather and air quality forecasting in tropical regions, especially in South America, the Regional Atmospheric Modeling System (RAMS) adapted for use in Brazil (Cotton et al., 2003; Freitas et al., 2009) was augmented with a module designated the Simplified Photochemical Module described in Freitas et al. (2005). This version of the model is being called the BRAMS-SPM and is running in an operational mode at the Institute of Astronomy, Geophysics and Atmospheric Sciences of the University of São Paulo (IAG-USP), with the goal of forecasting the concentrations of tropospheric ozone, nitrogen oxides, VOCs, sulfur dioxide, and small particles (PM₁₀ and PM_2.5) over the MASP, in a 48–72 h time frame. Considering its application over very large urban regions, the BRAMS-SPM is also capable of representing in a very consistent way the effects of the urban structure, as well as those of anthropogenic heat and moisture emissions, by using the Town Energy Budget parameterization, as initially proposed by Masson (2000) and first introduced into the RAMS model by Rozoff et al. (2003), in conjunction with the soil-vegetation-atmosphere transfer scheme, the Land Ecosystem-Atmosphere Feedback model, version 2 (Walko et al., 2000). This is a very important feature, because urban air quality is highly dependent on mesoscale circulations generated by anomalous urban heating and roughness, creating the so called urban heat island effect, mentioned in studies conducted in various countries (Oke, 1987; Yoshikado and Tsuchida, 1996; Ichinose et al., 1999; Baik et al., 2001; Cenedese and Monti, 2003; Gedzelman et al., 2003; Childs and Raman, 2005; Jonsson, 2005; Freitas et al., 2007; Roth, 2007). Urban traffic behavior is simulated by considering pollutants and anthropogenic fluxes (of heat and moisture). Rush hours are taking into account by superposing two normal distributions, each centered on the peak hours during the morning and the afternoon, respectively, as described by Freitas et al. (2007).

As proposed by Masson (2000), the model uses the urban canyon approach, in which a street or road is flanked by two facing walls that are similar in terms of height, width, and roughness, as well as thermal properties such as albedo, conductivity, and emissivity). Although the urban canyon approach uses the properties of the buildings in an urban area, the main differences found in urban areas are also considered in the model by defining four different urban types in Land Use/Land Class (LULC) images. Figure 2 shows the LULC image, adapted from Gouvêa (2007), for the MASP. As can be seen, the LULC image for the MASP defines four different urban or suburban types. Each of the urban types has a different structure. Table 1 shows the characteristics of the urban/suburban types defined for the MASP and used in the BRAMS-SPM model. The model is also equipped with important physical parameterizations, including turbulence; deep and shallow convection; microphysics; and soil-vegetation-atmosphere interaction. The main physical parameterizations frequently used during air quality simulations are presented in Table 2.

Figure 2. Land Use/Land Class image used in the BRAMS-SPM model for the Metropolitan Area of São Paulo. Adapted from Gouvêa (2007).

Table 1. Some values of Town Energy Budget parameters used in the BRAMS-SPM model to differentiate characteristics based on the urban canyon concept.

Table 2. Physical parameterizations used in the BRAMS-SPM.

WRF/Chem Modeling

The WRF/Chem is a non-hydrostatic mesoscale numerical weather prediction system designed for use in atmospheric research and operation forecasting. The system was developed through a collaborative effort among institutions and is available at http://www.mmm.ucar.edu/wrf/users (Skamarock et al., 2008). As described by Grell et al. (2005), the WRF/Chem has a meteorological component coupled with a chemistry module. This chemistry module is an atmospheric chemistry approach to representing gaseous, aerosol, and aqueous chemistry processes. Various user-selected combinations of chemistry schemes are available to represent the chemical and physical processes of gasses and aerosols, such as dry deposition; biogenic and anthropogenic emissions (from stack and vehicular sources); biomass burning; and aerosol schemes.

In WRF/Chem a preprocessing scheme (WPS), with the grid system and geographical data (i.e., topography, land use, ground cover, etc.) are defined and used in order to prepare the fields. For this task WRF/Chem uses the High Resolution Land Data Assimilation System (HRLDAS) which prepares surface initial fields in agreement with the grid domain and horizontal grid spacing. Details of this procedure can be found at http://www.ral.ucar.edu/research/land/technology/lsm/noah/hrldas/HRLDAS_USERS_GUIDE.pdf. The model is managed by a namelist type file (namelist.input), in which users can chose and setup the chemical and physical parameterizations, as well as the time step, simulation period, grid configuration, output time, and types of emissions (such as biogenic, anthropogenic, biomass burning, dust, and industrial stack emissions). The last step is the post processing, which can be done using the ARW program or a similar program in order to visualize the outputs. The computer system used is an SGI Altix XE 1300 Cluster. The system is composed by 8 computing nodes, each of them having 2 Intel Xeon 6 core 2—12 threads @ 3.07 GHz processors. During the simulations, we use 48 threads (2 nodes—4 processors) and the mean computational coast is approximately 13 min of processing for each simulated hour. In order to simulate 84 fours we spend about 18 h of processing.

The configurations that are being considered in the version of the model that is running operationally are presented in Table 3. The chemistry mechanism for the gas phase is the carbon-bond mechanism, version Z [CBM-Z, Zaveri and Peters (1999)], and the Model for Simulating Aerosol Interactions and Chemistry [MOSAIC, Zaveri et al. (2008)] is being used in the description of aerosols. The CBM-Z considers 67 prognostic species and 164 reactions with lumped hydrocarbons. The MOSAIC scheme considers the main aerosol species, including sulfate, nitrate, ammonium, sodium, calcium, black carbon, organic carbon, and liquid water with a sectional representation for the size distributions. This combination (CBMZ/8-bin MOSAIC) allows ethanol and other oxygenated compounds to be represented explicitly. Figure 3 presents the characteristic intervals for the size distribution of the fine and coarse aerosol considered in the MOSAIC scheme for the aerosols in the MASP, according to the data collected with a rotating cascade impactor and analyzed to provide the mass size distribution of the aerosol concentration and its composition in metals, sulfate, nitrate, ammonium, black carbon, organic carbon, and other compounds that can be explicitly represented in the model.

Table 3. Model configuration options in the WRF/Chem simulations.

Figure 3. Proportion of the aerosol mass distributed at each size interval considered in the MOSAIC approach applied to the simulations.

Data Sources

Anthropogenic Emissions

The main source of atmospheric pollutants in the metropolitan areas of southeastern Brazil is vehicle emissions. In this study, we collected information from CETESB, from the IAG-USP Laboratory of Atmospheric Processes, and from DENATRAN. Emission factors for different vehicle types (motorcycles, light-duty vehicles, and heavy-duty vehicles) and different fuel types (gasohol, ethanol, combination of any proportion of gasohol and ethanol, and diesel), as well as intensity of use were obtained from CETESB reports (CETESB, 2012). Studies conducted at the IAG-USP Laboratory of Atmospheric Processes provided emission factors estimated from in-tunnel field campaigns, as well as temporal distribution (Martins et al., 2006; Sánchez-Ccoyllo et al., 2009; Nogueira et al., 2014; Pérez-martínez et al., 2014). Information on the numbers of vehicles in the different categories is available on the DENATRAN website (http://www.denatran.gov.br/frota.htm). For the spatial distribution, we assumed that the vehicles within the modeling domain were distributed proportional to the road length in each grid cell. Road length was calculated as the sum of various types of road within each grid cell. The road map is available on the OpenStreetMap website (http://www.openstreetmap.org). Initially, the total number of vehicles was calculated, on the basis of DENATRAN data for 2014, for each state within the modeling domain (Table 4). The proportional distribution of vehicles, by fuel type, is shown in Table 5. For each type of vehicle/fuel combination, an average intensity of use was estimated from CETESB data (CETESB, 2012), as shown in Table 6.

Table 4. Total number of vehicles in each of the states within the modeling domain.

Table 5. Vehicle types and proportional use of fuels.

Table 6. Intensity of use, by vehicle type.

Emission factors were preferentially taken from the tunnel experiments. However, when those were not available, CETESB (2010) emission factors were used if they corresponded to the same period evaluated in the tunnel experiments. Figure 4 shows the emissions factors for CO, NO_x, SO₂, and PM₁₀, whereas Figure 5 shows the VOC emission factors related to exhaust, evaporative, and liquid emissions for vehicles running with gasohol and flex fuel vehicles (as they can run with ethanol and gasohol). The VOC speciation was made considering measurements performed in tunnels and with dynamometers.

Figure 4. Emission factors for CO, NO_x, SO₂, and particles for different vehicle type/fuel combinations. LD-G, light-duty vehicles burning gasohol; LD-F, light-duty vehicles flex-fuel; Moto-G, motorcycles burning gasohol; Moto-F, motorcycles flex-fuel.

Figure 5. VOC emission factors from exhaust, evaporative and liquid emissions. Exh, exhaust; Eva, evaporative; Liq, liquid; LD-G, light-duty vehicles burning gasohol; LD-F, light-duty vehicles flex-fuel; Moto-G, motorcycles burning gasohol; Moto-F, motorcycles flex-fuel.

Figure 6 presents the relative fractions of VOC emissions from different processes and fuels. These fractions are the proportions emitted considering the VOC emission, which is calculated based on the emission factors presented in Figure 4. Because large amounts of ethanol are consumed in Brazil, we represent explicitly the emission of ethanol, formaldehyde, and acetaldehyde from exhaust. The values used were 0.25 g/km, 0.01 g/km, and 0.03 g/km, respectively, for vehicles burning ethanol. We assumed that all the grid cells would have the same temporal distribution of emissions. Figure 7 shows the normalized average vehicles counts from the tunnel experiments. Finally, road length (the sum of motorway, trunk, primary, secondary, and tertiary types of roads) was calculated for each grid cell, as shown in Figure 8.

Figure 6. VOC fractions for exhaust, evaporative, and liquid emissions from gasohol (A), ethanol (B), and diesel (C). G, gasohol; E, ethanol; D, diesel; -vap, evaporative; -liq, liquid; -exh, exhaust; e_eth, ethane and alkanes with kOH < 0.5 × 10³ ppm⁻¹ min⁻¹; e_hc3, propane and alkanes with kOH 0.5 × 10³–5.0 × 10³ ppm⁻¹ min⁻¹; e_hc5, alkanes with kOH 5.0 × 10³–1.0 × 10⁴ ppm⁻¹ min⁻¹; e_hc8, alkanes with kOH > 1.0 × 10⁴ ppm⁻¹ min⁻¹; e_ol2, ethylene; e_olt, propylene and alkenes with <2.0 × 10⁴ ppm⁻¹ min⁻¹; _e_oli, dienes and alkenes with >2.0 × 10⁴ ppm⁻¹ min⁻¹; e_isop, isoprene; e_tol, benzene and aromatics with <2.0 × 10⁴ ppm⁻¹ min⁻¹; e_xly, xylenes and aromatics with >2.0 × 10⁴ ppm⁻¹ min⁻¹; e_ket, ketones; e_c2h5oh, ethanol; e_hcho, formaldehyde; e_ald, acetaldehyde and other aldehydes; e_ch3oh, methanol.

Figure 7. Temporal distribution of emissions for light-duty vehicles (LDV, including motorcycles) and heavy-duty vehicles (HDV).

Figure 8. Road length (in km) for each 9 × 9 km grid cell.

Air Quality and Meteorological Data

The modeling is being evaluated by comparing experimental data (meteorological parameters and pollutant concentrations, measured at different points in the area) and simulation data. The air quality data that is being considered for comparison in the MASP is provided by the CETESB air quality network. The location of the air quality stations and the parameters measured are presented in Table 7. Other information regarding the measurement of hydrocarbons, ozone precursors, and particles are from experimental campaigns carried out at different sites, mainly in the MASP, and from the IAG-USP Laboratory of Atmospheric Processes. To evaluate particle size distributions for mass, number distribution of fine particles, and cloud condensation nuclei, a micro-orifice uniform deposit impactor was employed (Ynoue and Andrade, 2004), whereas particle concentrations and composition were evaluated with receptor modeling (Andrade et al., 2012b; Miranda et al., 2012).

Table 7. Surface data available for evaluation of the modeling results.

To study the vertical profile of ozone, ozonesondes have being launched in the downtown area of the city of São Paulo. There were two experimental campaigns, one conducted in 2006 (Andrade et al., 2012a), with eight soundings in the fall (from May 15 to May 18) and eight in the spring (from October 29 to November 1), and another conducted in 2013 (Andrade et al., 2014), 10 ozonesondes being launched between October 29 and November 4. The parameters analyzed were ozone concentration, wind direction, wind velocity, and temperature. Simulations with WRF-Chem for the period when the ozonesondes were launched, from October 29 to November 4 were performed in order to determine the quality of the simulations by comparing the modeled and measured data. Figure 9 shows the comparison between modeling and measurement for ozone concentration and temperature, on November 2. The major differences were in ozone concentration near the surface, which could be associated with the great uncertainty in the emissions inventory. One important point was the presence of clouds affecting ozone formation near the surface on that day.

Figure 9. Comparison between sounding and modeling for ozone (A) and temperature (B), on November 2, 2013.

Results from the WRF-Chem Model

Hourly data collected from 2011 until the present have been compared with the measurements of NO_x, CO, ozone, and particles, to evaluate the performance of the model. The incorporation of better descriptions of the emission profiles and a representation of the atmospheric chemistry have made it possible to provide a forecast (www.lapat.iag.usp.br/). Some results of a recent case study are presented in order to illustrate the importance of forecasting the behavior of pollutants.

Case Study: Summer 2013/2014

The comparison between measurement and modeling is presented for the period of January–February 2014. The summer of 2013/2014 was chosen because it was a period when there was considerably less precipitation and humidity than would be expected on the basis of the historical climatological data (for 1933–2013). During the summer months of December 2013, January 2014, and February 2014, the total rainfall was 352.5 mm, compared with the climatological mean of 634.8 mm. Total rainfall was 199.3 mm in January and 81.1 mm in February, compared with the climatological means of 231.9 mm and 212.5 mm, respectively. In the summer of 2013/2014, the temperature was also above the climatological mean. In February, the relative humidity was below 40% on 17 days, whereas the historical mean is 2 days. During the summer of 2013/2014 as a whole, there were 27 days on which the relative humidity was below 40%.

The quality of the simulation, in comparison with direct measurement, was evaluated with the statistical parameters correlation, bias, root mean square decomposed into two components: systematic RMSDs and unsystematic RMSDu, and agreement index, according to Willmott et al. (1985) and Lu et al. (1997). The statistical parameters for ozone, NO_x, and PM_2.5 are presented in Table 8 for the station located at the Institute for Energy Research and Nuclear Science on the campus of the University of São Paulo, IPEN station (23°33'S; 46°44'W). The strongest and weakest correlations found were for ozone and PM_2.5, respectively. Figure 10 shows the comparison between the measured and simulated values (with the standard deviation) for O₃, NO_x, and PM_2.5 for the January–February 2014 period for the IPEN station. The high concentrations during that period are associated with the favorable atmospheric conditions of less precipitation and high temperatures. The panels (A, B, C) illustrate the concentration and standard deviation of measured and simulated O₃, NO_x, and PM_2.5, respectively, panels (D, E, F) illustrate the Root-Mean-Square-Deviation (RMSD) separate in systematic (RMSDs) and unsystematic (RMSDu) for the same pollutants, and (G, H, I) illustrate the agreement index for the studied pollutants. The RMSDs, the linear bias, explains most of the RMSD for the PM_2.5, while for O₃ the RMSDu, the accuracy of the simulations is lower. The agreement indices indicate that the ozone pattern simulations in the afternoon are in good agreement with observations, but for NO_x and PM2.5 the model is not accurately representing the concentrations measured.

Table 8. Statistical parameters indicating the quality of the simulation for the period of January–February 2014.

Figure 10. Comparison between modeled and simulated values for O₃, NO_x, and PM_2.5 during the January–February 2014 period for the air quality monitoring station at the Instituto de Pesquisas Energéticas e Nucleares (IPEN, Institute for Energy Research and Nuclear Science) on the campus of the University of São Paulo. (A–C) Illustrate the concentration and standard deviation of measured and simulated O₃, NO_x, and PM_2.5, (D–F) illustrate the Root-Mean-Square-Deviation (RMSD) separate in systematic and unsystematic, and (G–I) illustrate the agreement index for the studied pollutants.

The spatial distribution of the pollutants illustrates the role that wind plays in the transport and position of the highest concentrations of the secondary compounds. Figure 11 presents the concentrations of NO_x and ozone at 6:00 a.m., 12:00 p.m., and 6:00 p.m. on January 3 for the entire grid. The metropolitan areas are sources of pollutants that affect the surrounding areas according to the topography and meteorological systems. The coast of the southeastern coast of Brazil region induces the sea-breeze circulation, and the proximity with mountains to the northwest trap the plume in that direction. The highest concentration of ozone was to the southeast, induced by the strong solar radiation during the period and wind circulation to the southeast, which is in agreement with patterns of (mesoscale) wind circulation at this time.

Figure 11. Spatial distribution of NO_x (left side) and ozone (right side), for the modeling area as a whole, at 6:00 a.m. (top), 12:00 p.m. (middle), and 6:00 p.m. (bottom) on January 3, 2014.

Results from the BRAMS-SPM Model

In recent years, the BRAMS-SPM model has been applied in a large number of studies conducted in Brazil, most focusing on the atmosphere over the MASP, over the Metropolitan Area of Rio de Janeiro, or over the central region of the country (Vendrasco et al., 2009). For example, Carvalho (2010) applied the BRAMS-SPM model for the study of ozone concentrations near the surface over the Metropolitan Area of Rio de Janeiro during the summer and autumn of 2007. Industrial and vehicle emissions during the simulations were provided by the Rio de Janeiro State Environmental Agency, and their spatial distribution is shown in Figure 12. The vehicular emissions input files in the SPM-BRAMS model were built based in the urban areas identification and emission data available. Industrial emissions were added through a text file addressed in the model namelist (RAMSIN File).

Figure 12. Spatial distribution of industrial emissions of SO₂ (A) and vehicle emissions of CO (B) over the Metropolitan Area of Rio de Janeiro in 2007. Sources: Rio de Janeiro State Environmental Agency; Carvalho (2010).

The diurnal cycle of vehicle emissions was based on the mean cycle of carbon oxide emissions between 2000 and 2008. Figure 13, obtained from Carvalho (2010), shows very good agreement between the observed and modeled values for ozone over a 10-day simulation period.

Figure 13. BRAMS-SPM-simulated values for ozone over two sites in the Metropolitan Area of Rio de Janeiro between March 5 and March 14 2007. Sources: Rio de Janeiro Instituto Estadual do Ambiente (INEA), State Environmental Agency; Carvalho (2010).

Discussion

We have presented operational tools, developed using the WRF/Chem and BRAMS-SPM models as platforms, for forecasting pollutant concentrations in the southeastern region of Brazil and for studying the photochemical formation of pollutants and secondary particles. Major efforts have been applied to describe the emissions inventory considering the real conditions of vehicular traffic in the metropolitan areas of the region. To that end, we applied a novel scheme using data from traffic counts and road distributions. With this recent implementation, the performance of the WRF/Chem model has improved. The statistical tests applied to compare simulation and measurement showed good quality for ozone but that it is necessary to improve the representation of the NO_x and fine particles; the correlation between measured and simulated ozone concentrations was 0.84 with a bias of –13.7, showing that the model is representing the behavior of the pollutant but not the highest values. For fine particles the model is estimating average concentrations 10 μg/m³ higher than the measured values. NO_x is a primary pollutant very difficult to evaluate due to the impact of local emissions from mobile sources to the concentrations measured at the air quality stations. With this performance, the modeling system can be applied to evaluate public policies concerning fuels and the distribution of urban sources of emissions. In order to achieve this objective the emission inventory module must be continuously updated with emissions factors and activity information for the mobile source. The uncertainties in the boundary conditions constitute important aspects that have to be correct evaluated, specifically to correctly represent the vertical profiles of the studied pollutants. Measurements of particulate matter size distributions of concentrations and compositions are important for the description of microphysics process of cloud formation and radiation scattering and absorption.

Conclusions

The forecasts of pollutant concentrations can be accessed by the community and authorities interested in air quality and its effects on public health. The implementation of the modeling system with an emphasis on vehicle emissions is important for the metropolitan areas in Brazil, because such emissions constitute a major source of air contamination. Consumer switching between ethanol and gasohol in response to fluctuations in their prices underscores the need for an accurate representation of the chemistry of the gas phase, including oxygenated compounds. In the metropolitan areas of Brazil, concentrations of fine particles and ozone are major concerns, which call for the ongoing improvement of the emission scheme. The emission scheme approximation for the spatial distribution of the mobile source developed in this study can be applied to other regions that would like to develop air quality modeling but don't have available the information needed to construct the mobile emission field.

Conflict of Interest Statement

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Acknowledgments

The authors thank the CAPES (Coordenação de Aperfeiçoamento de Pessoal de Nível Superior) and FAPESP (São Paulo State Research Foundation), Research Program on Global Climate Change, for the financial support of this work (Processes 2008/58104-8). We also thank the State Company for the Environment CETESB. Finally the authors thank Mr. Jefferson Boyles for the revision of the text.

References