Experimental and model assessment of PM 2.5 and BC emissions and concentrations in a Brazilian city – the Curitiba case study

Data on airborne fine particle emissions and conce ntrations in cities are valuable to traffic and air quality managers, urban planners and landscape architects, health practitioners, researchers, and ultimately t o legislators and decision makers. This study aimed at determining th e emissions and ambient concentrations of black car bon (BC) and fine particles (PM2.5) in the city of Curitiba, southern Brazil. The applied methodology applied combined a month-long monitoring campaign that included both fixed and mo bile instruments, the development of emission inven tories, and the 20 dispersion simulation of dispersion from the regional down to the street scale. The mean urban background PM 2.5 concentrations during the campaign were below 10 μ g m -3 in Curitiba city center, but twoto three-fold higher in a residential area in the city outskirts , indicating the presence of unidentified local sou rces, possibly linked to wood combustion. Mean BC concent rations seemed to be more uniformly distributed ove r th city, with urban background levels around 2 μg m , which rose to about 5 μg m -3 in heavily trafficked street canyons. The dispersi on 25 modeling also showed high PM 2.5 and BC concentrations along the heavily transited ring road and over the industrial area southwest of Curitiba. However, the lack of in situ data over this area prevented the corroboration of the model outputs. The integrated approach used in this study can be imple ented in other Brazilian cities as long as an open data policy and a close cooperation between among municipal authorities and academia can be achieved.

industrialization have increased the concentrations of outdoor air pollutantsion from vehicles and factories, contributing to the global burden of asthma and allergic diseases (Zhang et al., 2015). A recent global model assessment with relatively high spatial resolution (11x11 km 2 ) revealed that the human exposure to ambient PM 2.5 , i.e., fine particles with an aerodynamic diameter smaller and or equal to 2.5 µm, ranked fifth as mortality risk factor in 2015 with more than four million deaths (Cohen et al., 2017). The urgent need to take better care of cities' environment has recently been manifested in the 5 Sustainable Development Goal (SDG) 11 addressing urban planning and development. One specific SDG indicator, 11.6.2, for meeting this goal is the annual mean level of PM 2.5 weighted by population (United Nations Statistics Division, 2018), for which a global database was developed by the World Health Organization (WHO, 2016). Improved information on the spatial distribution of PM 2.5 exposure in cities will be required to monitor future progress of the SDG 11 air pollution target.
Ongoing research is trying to identify which components of inhalable particulate matter contribute the most to the observed 10 health effects. The Review of Evidence on Health Aspects of Air Pollution (REVIHAAP) assessment (WHO, 2013) could not pinpoint the constituents within the PM 2.5 matrix that are more related to specific health outcomes. However, strong research outcomes have shown that black carbon (BC) particles -associated to combustion emissions, especially from diesel engines-are a more robust indicator of health effects than solely PM 2.5 (Janssen, et al., 2011). An additional aspect of BC as a metric for air pollution is its potential to act as a Short-Lived Climate Forcer (SLCF), contributing to warm up the 15 atmosphere (Bond et al., 2013). Hence, curbing BC emissions has the double benefit of reducing the human exposure to PM 2.5 , while contributing to mitigate climate change.
Global assessments of air pollution have revealed that the highest PM 2.5 exposure levels are found in Asia, whereas lower exposure occurs in the American continent (WHO, 2016). However, there is no threshold for the PM 2.5 health effects below which the consequences for human health are negligible. On the contrary, there are indications that a reduction of 1 µg m −3 20 on PM 2.5 concentrations provides a more significant health benefit in a city with relatively low levels (i.e., in the range of 10-30 µg m −3 ), as compared to cities with much higher mean concentrations, close or above 100 µg m -3 (Burnett et al., 2014).
Another aspect, illustrating the importance of reducing BC, is exemplified by Janssen et al. (2011), who showed that a 1 µg m −3 decrease in PM 2.5 exposure would lead to an increase in life expectancy of 21 days per person, whereas the same reduction in BC concentration would yield an increase between 3.1 and 4.5 months. 25 The national ambient air quality standards (NAAQS) in Brazil were implemented in 1990 and legislate particulate matter concentrations as total suspended particles, PM 10  The official air quality monitoring network is managed by the Environmental Institute of Paraná (IAP) and consists of four automatic stations within the municipality of Curitiba: Boqueirão (BOQ), Cidade Industrial (CIC), Ouvidor Pardinho (PAR), Santa Cândida (STC) and four in the industrial area of the nearby city of Araucária. An analysis of PM 10 and NO 2 data from the official monitoring network was performed for the three year period 2013 to 2015. The objective was to describe the general pollution levels, extreme values and seasonal variability inside the Curitiba municipality, for a period close to the 5 campaign performed as part of this project in 2016, i.e., for conditions with similar urban emissions.
The air quality in Curitiba is affected by emissions generated inside the city and by sources located outside in the Paraná state and in other parts the continent. An observational and modelling study conducted by Rosário et al. (2013) showed that smoke plumes generated during the biomass burning season in northern and central Brazil (August-October with and peak activity in September) reached the southern states, including the Curitiba region. 10

Study design
A combination of measurements and dispersion modeling was used to assess and map PM 2.5 and BC concentrations in Curitiba. The modeling had two purposes: validating the emission inventory through comparisons with measured data at a few stations, and obtaining a spatial distribution of the PM 2.5 and BC concentrations over the city. The comparison of 15 monitored data and model output can be used for microenvironments impacted by one dominating source to allow an in situ determination of the emission factors, in this case for vehicles driven in the city.
The full assessment included the following components (see map in Fig. 1): 1. Analysing the NO 2 /NO x and PM 10 concentrations collected between 2013 and 2015 at the four IAP official monitoring sites in Curitiba: PAR, BOQ, STC and CIC (the monitoring network also includes stations outside Curitiba, however 20 they were not used for comparisons with model output since the high resolution modelling only covered the Curitiba municipality).
2. Developing an emission inventory for the city and the state of Paraná.
3. Performing field campaign at two fixed sites ( Fig. 1 left) aimed at: a) monitoring of NO x , PM 2.5 and BC concentrations within a street canyon (Marechal Deodoro, hereafter MD) in the 25 city center at two levels above ground: street (height of 5 m, hereafter 'MD street') and rooftop (height of 70 m, hereafter 'MD roof'). b) monitoring of PM 2.5 , BC, EC (elemental carbon) and OC (organic carbon) concentrations in a residential area (Sítio Cercado, hereafter SC) located 13 km from the city center, and 750 m from the heavily trafficked road BR-376, which is part of Curitiba's ring road ('Contorno' as local designation). 30 4. Performing a monitoring campaign with instruments on-board bicycles to measure PM 2.5 and BC concentrations along different types of roads in the city center (see Fig. 1, right). 5. Implementing dispersion models at the regional, urban and street canyon scales to support the interpretation of the monitored data inside the Curitiba municipality.
6. Consolidating the street canyon data and model output to obtain real-world emission factors for PM 2.5 and BC for road transport in Curitiba. 7. Using the regional and urban modeling, together with the monitored data, to conduct a source apportionment of PM 2. 5 5 and BC levels in the Curitiba municipality.
The modeling in a microenvironment -a street canyon -had the purpose to determine the impact on air quality levels by one dominating source, to allow an in situ determination of the emission factors, in this case for vehicles circulating in the city center. The comparison between measurements and model output was also extended to the urban background of the Curitiba municipality, comparing model results with the measurements in the city center and in a residential area, this in order to 10 evaluate the emission inventory covering the Curitiba municipality, i.e., the area for which high resolution modelling was performed.

Emission inventory
The emission inventory developed for Curitiba considered PM 2.5 , BC and NO x for two major economic sectors: industries 15 and on-road transport. An attempt was made to collect data on the use of wood or coal by restaurants. However, the database gathered was incomplete in space, impeding the inclusion of this source in the emission inventory. Neither was it possible to obtain data on the residential use of wood stoves for cooking or heating.; hHowever, municipal authorities informed that residential wood combustion should be minimal, at least in the city center.
Industrial emission values of NO x , SO 2 and PM 10 from large industrial sources were compiled from the official regional 20 inventory that covers the state of Paraná (IAP, 2013), while the Curitiba municipality provided data for the inner-city small scale industries. Because Since these industrial inventories only included PM 10 emissions, we assumed that 70% of the PM 10 emitted by the industries in and around Curitiba consisted of PM 2.5 (Erlich et al., 2007). Information on industrial BC emissions was not available. All industrial emissions were treated as point sources with emissions coming out of a stack with characteristics given by the IAP inventory. The IAP inventory is based on a policy in which the industries are supposed to 25 monitor themselves their own emissions and communicate to the official authorities. The IAP inventory reveals a concentration of sources in the southwest area of Curitiba, and in the neighbor city Araucária. Within the central parts of Curitiba there are just a few point sources, due to the urban planning of the 1970's and the creation of the Curitiba's Industrial Site located in the southwest region, moving almost all industries in the center to this new area downwind of the city according to the predominating winds. The Araucária city includes one of the biggest industrial sites of Brazil with a 30 state-owned oil refinery together with several steel, cellulose/pulp and chemical industries.
Traffic emissions in Paraná state, but outside the city of Curitiba, were also extracted from the official regional inventory (IAP, 2013). Within the city, the emission calculations were split between public transport (buses) and private vehicle emissions. Public transport emissions were calculated as line sources along the bus network, as facilitated by the municipality. Information on bus size, technology (Table 1), average daily distance travelled and average fuel consumed were obtained for each bus line. Together with the bus timetables, it was possible to describe the number of buses transiting a certain road link on an hourly basis. We used a relation between the Brazilian emission legislation PROCONVE and the European EURO classes (TransportPolicy.net, 2018a; 2018b), enabling the use of emission factors from the European 5 HBEFA database (INFRAS, 2017), as listed in Table 2. We assumed an average bus speed of 50 km h -1 and saturated traffic conditions. The bus types listed in Table 1 were aggregated into HBEFA bus classes with weights <15 ton, 15-18 ton, and >18 ton, respectively. Emission factors for bi-articulated buses were extrapolated using information on fuel consumption provided by the Curitiba municipality. BC emission factors were taken from the European Environment Agency EEA (EEA, 2016) as BC/PM 2.5 fractions in the range 65-75%. The PM and BC emission factors for buses useing of 100% biofuel lowers 10 the PM and BC emissions were reduced by 50%, according to the findings published by the U.S. Department of Energy  (Fig. 2). The same daily profiles were used for all streets.
As for the fleet composition, a simplified assumption was made with three different compositions depending on the type of road. The following shares were adopted, as suggested by the traffic engineers at the Curitiba Municipality: for inner-city 25 roads (as limited by the ring road): 93% cars, 5% utility vehicles and 2% trucks; for a few larger thoroughfares open for heavy-duty vehicles: 82% cars, 8% utility and 10% trucks; and for the ring road: 59% cars, 9% utility and 32% trucks. Table   3 shows the assumptions made on size and technology, together with emission factors taken from EEA's emission inventory guidebook the European Environment Agency (EEA, 2016). Utility vehicles and trucks were assumed to be diesel-fueled, while flex-fuel cars were running on gasoline as drivers' primary fuel choice due to favorable price. Stop-and-go emissions 30 due to congestion or traffic lights were not considered. It follows from the emission factors displayed in Table 3 that the most critical assumption is the percentages of diesel fueled LDVs and HDVs.
An attempt was made to collect data on the use of wood or coal by restaurants. However, the database gathered was incomplete in space, impeding the inclusion of this source in the emission inventory. Neither was it possible to obtain data on the residential use of wood stoves for cooking or heating. However, municipal authorities informed that residential wood combustion should be minimal, at least in the city center.

Monitoring campaign
Fixed-site measurements were conducted in the winter period July 25-August 24, 2016, when pollution levels are expected to peak (see Section 3.1 for further details). Mobile monitoring was performed on 10 selected days and times (morning and 5 evening rush hours) between August 1 and August 14, 2016. PM 2.5 monitoring was conducted with three types of instruments: a Harvard Impactor using cut-off 2.5 microns (deployed at street and rooftop levels of the canyon site), a MicroVol low volume air sampler (model 1100, Ecotech, Australia) at SC, and three DustTrak units (model 8520, TSI, USA), one deployed at SC and two units used for mobile measurements.
The Harvard Impactor collected PM 2.5 samples on 37-mm teflon filters for gravimetric analysis with a 24-hour resolution. 10 Daily integrated samples for gravimetric, EC and OC analysis were collected on 47-mm quartz fiber filters using the Ecotech MicroVol. The gravimetric analyses were performed following the NIOSH Method 5000 (NIOSH, 2003), while the EC and OC analyses were conducted at Stockholm University using the NIOSH temperature protocol (Birch, 2003). The PM 2.5 output from the DustTrak instruments was calibrated with the gravimetric data from SC, yielding a correction factor of 1.92 (R 2 = 0.77), which was subsequently applied to correct all DustTrak outputs. This correction factor is within ranges reported 15 by other studies conducted in urban areas, from 1.70 (McNamara et al., 2011) to 2.78 (Wallace et al., 2011).
Total BC concentrations at street and roof levels were measured with aethalometers (models AE42 and AE33, respectively, Magee Scientific, USA) operating at seven wavelengths (370, 470, 520, 590, 660, 880, and 950 nm) and with five minute time resolution. At SC and for the mobile measurements, BC concentrations were determined with microaethalometers (model AE51, AethLabs, USA) operating at the wavelength of 880 nm. These instruments use the wavelength-dependent 20 absorption cross-section values provided by the instrument manufacturers to convert aerosol absorption coefficient into BC mass concentrations. In this study, a site-specific absorption cross section of 18.39 m 2 g -1 was determined at SC by correlating daily mean aerosol absorption coefficients with collocated EC concentrations (R 2 = 0.96). The BC data from the AE33 and AE42 instruments were determined using the absorption cross section provided by the manufacturer.
To supplement the datasets and support the modelling activities, high spatio-temporal BC and PM 2.5 concentrations were 25 collected on selected days and hours along prescribed routes using two retrofitted bicycles as sampling platforms, following the experimental design by .
The passive sampling of NO x at the roof and street level of the MD site was performed using Ogawa passive samplers (Hagenbjörk-Gustafsson et al., 2010) during two periods of 14 days each. The purpose of this NO x measurement was to support the determination of local emission factors for vehicle emitted PM 2.5 and BC. 8

Dispersion modeling
Dispersion modeling was performed at three spatial scales with three different models, all giving hourly data for the monitoring period from July 25 to August 24, 2016. The regional scale modeling was performed with the BRAMS 5.2 modeling system (Freitas et al., 2017), which includes an atmospheric chemistry transport model (CCATT) coupled on-line with a limited-area atmospheric model. For this specific experiment, the model was configured to simulate aerosol emission, 5 transport and its effects (dispersion option) during the campaign period. The BRAMS physical parameterizations were configured with Mellor-Yamada level-2.5 turbulence scheme (Mellor and Yamada, 1982) and Joint UK Land Environment Simulator (JULES) surface-atmosphere interaction model (Moreira et al., 2013). To For shortwave and longwave radiation schemes was used the Rapid Radiative Transfer Model for General Circulation Models (RRTMG) was used with 1200s frequency update of the radiation trend (Iacono et al., 2008). Finally the Grell and Freitas (2014) ensemble version for deep 10 and shallow convection and the single-moment bulk microphysics parameterization from Walko et al. (1995) were used. The model was used applied over two domains: one covering large parts of South America with a gridspatial resolution of 50x50 km 2 (G1, Fig. 32, left) and nested down to a 10x10 km 2 grid resolution over the state of Paraná (G2, Fig. 32, left) with anthropogenic emissions taken from a South American inventory (Alonso et al., 2010). The emissions for the state of Paraná were updated with the vehicular and industrial inventory described earlier (IAP, 2013), except for BC and OC that usedfor 15 which information from EDGAR-HTAP (Joint Research Centre, 2018) was used. The carbonaceous speciation BC and OC, from the HTAP database, is for South America the same as of Edgar V4.1 database, both at country and region level.
Biomass burning sources were taken from the model 3BEM (Longo et al., 2009) and biogenic emissions of gases from MEGAN (Guenther et al., 2012). These emission fields were generated by the preprocessor PREP-CHEM-SRC (Freitas et al., 2011). 20 For the urban scale modeling, a Gaussian dispersion model was used over a 32x32 km 2 domain (Fig. 32, right) with a grid spacing spatial resolution of 200x200 m 2 . This model is part of the Airviro system (Airviro, 2018) and incorporates a diagnostic wind model (Danard, 1977) that takes into account surface roughness and calculated over built-up areas as a function of building heights, with a maximum roughness of 0.7 m for the city center areas with the highest buildings. so that the model output over areas with high buildings will reflect only the conditions at roof level (i.e., the urban background). The 25 wind model assumes that small scale winds can be seen as a local adaptation of large scale winds (free winds) due to local fluxes of heat and momentum at the surface. The free wind is estimated from a vertical profile at the location of a meteorological station, using scaled stability variables. For this application, the input meteorological data was measured at one location (MET station in Fig. 1, left) and the building heights are illustrated in Fig. 32, right. The regional model output could be directly added to the urban model without double-counting, since the sources within the Curitiba municipality were 30 excluded in the regional model.
The street canyon simulations at the MD monitoring site were performed with the OSPM model (Berkowicz, 2000), one of the models available in the Airviro system. This street canyon model consists of two components: a plume model that simulates the direct contribution from vehicles and a box model that calculates the contribution from the recirculation part created by the vortex. This street canyon model consists of two parts, one direct plume model, following the estimated wind direction at the bottom of the street canyon. The second model part takes care of the contribution from the vortex-like recirculation created by the surrounded buildings and is calculated by a simple box model. Wind input data are taken from the diagnostic wind model, assuming its output to represent the wind at roof level. Neutral stability is assumed within the street 5 canyon. The OSPM model was only used to determine the air pollution contribution from local traffic inside the MD street canyon, to be compared to the measured increment from the roof to street level. Building heights and the dimensions of the street canyons were determined using 3D Lidar data provided by the Curitiba municipality. Private traffic volumes passing this station were estimated by municipal traffic experts as 24,075 vehicles per day, as an average over the week, composed by 93% of cars, 5% light duty diesel vehicles and 2% heavy duty diesel trucks. The public transport consisted of 469 buses 10 per day.

PM 10 and NO 2 concentrations as registered by the official monitoring network
The highest PM 10 concentrations during the three year period between 2013 and 2015 were recorded at station CIC (Table 4, note the low data capture at this station), located close to the industrial area and the ring road, with both mean and maximum 15 daily values within NAAQS for Brazil (50 and 150 µg m -3 , respectively). The same applies to the NO 2 annual mean, for which the limit value is 100 µg m -3 and the maximum measured annual mean was 26.5 µg m -3 at station PAR, located at a square in the city center. Averaginged over the three years, all stations showed the highest values in August, which incentivized the PM 2.5 and BC monitoring campaign of this study to be performed between July and August 2016.
The ratios of the mean PM 10 and NO 2 levels monitored at the official network during the monitoring campaign (inJuly 25 -20 August 24, 2016) to their respective mean values in August of 2013, 2014 and 2015 were 70% for PM 10 at PAR, 50% for PM 10 at BOQ, and 47% for NO 2 at PAR and 86% for NO 2 at STC., Unfortunately thewhile there were no PM 10 monitor atdata reported from station CIC was not in operation during the campaign period, however the NO 2 ratio at CIC was 76%.. These listed ratios show considerably lower pollution levels during the field campaign, as compared to the average for this period during the previous three years. 25 While the number of public buses in operation decreased with by 6% between 2013 and 2016, the official number of registered private vehicles in Curitiba shows an increase by of 4% (http://www.detran.pr.gov.br). Although the registered fleet does not necessarily equal the actual number of vehicles in circulation, it gives the best estimate of the private traffic tendency. Relating the industrial emissions, and despite the lack of data that could sustain an analysis of the trend, to the authors' knowledge no significant changes occurred over the reported 4-year period. Therefore local emissions in Curitiba 30 are likely not to have changed significantly from the period 2013-2015 to 2016 and the identified differences in pollution levels can most likely be attributed to variations in meteorological conditions and the long-range transported pollution arriving to Curitiba. A comparison of meteorological conditions showed considerably more precipitation during August 2016 (163 mm) compared to what is expected at this time of the year based on a 30-year climatology (73 mm), while temperature, wind direction and speed were similar to what has beenwas registered during in the last 10-20 years (INMET, 20198).

BC, PM 2.5 and NO x concentrations measured during the 2016 campaign
On average, BC concentrations were the highest at street level (5.5 µg m -3 ), followed by rooftop level (2.3 µg m -3 , 5 representing urban background conditions), and 2.2 µg m -3 at SC site. The mean street canyon increment, which is defined as the difference between street and roof levels, equaled 3.2 µg m -3 and can be attributed to local traffic along the street canyon.   Fig. 65 shows the strong co-variation of BC levels (r = 0.68 for hourly data) registered 13 km apart and also the peak values occurring both on weekdays and weekends. Note the high BC levels occurring during the two weekends of 6-7 and 13-14 August, not found during the previous weekend 30-31 July. After analyzing the available air pollutant data for the two high peak weekends, as well as 15 wind direction and speed, it seems as that it was the very low wind speed that made it possible for BC and PM emissions taking place inside or just outside the city, to be advected in high concentrations over the central (MD) and southern (SC) parts of the city. Although the available information does not allow the determination of the source (or the sources) of the observed high concentrations, we can rule out the local traffic -which can be assumed to be a major source contributor to BC levels during weekdays, but not during weekend nights -and long-range transport events from remote sources -which 20 should raise the pollution levels in all stations in a more homogeneous way -as the main reason to the high peak values during the two weekends. Filter sampling of daily PM 2.5 took place both at MD rooftop and street levels. At rooftop level, the mean PM 2.5 value over the campaign period was 7.3 µg m -3 . However, the data measured at street level were affected by a technical failure and could not be used. During the mobile monitoring with the bikebicycles, two 15-min records from weekdays' morning and 30 afternoon rush hours were obtained during stops at the MD street level station. Those data indicated a mean PM 2.5 to BC ratio of 3. Using this scale factor on the BC data obtained at street level (mean concentration of 5.5 µg m -3 over the period), a PM 2.5 concentration of 16.5 µg m -3 was estimated for the street level.
At the residential site SC, the gravimetric analysis of 17 filters yielded a mean daily PM 2.5 concentrations of 36.2 µg m -3 , of which 11.2 µg m -3 was OC and 2.6 µg m -3 EC. This means that the total carbon (EC+OC) contributed with 38% to the PM 2.5 mass, and yielded a mean OC/EC ratio of 4.4. The mean hourly PM 2.5 concentration as registered hour-by-hout by the DustTrak monitor during 25 days, after calibration against the 17 filter samples, was 25.3 µg m -3 . The longer time period for the PM 2.5 mean value obtained for the DustTrak monitor makes it more representative than the filter mean. 5 The NO x concentrations at MD street site were 43 and 55 µg m -3 at roof level, and 105 and 122 µg m -3 at street level, for the two 14-day periods of the passive samplers. From this, the street canyon increment was determined as 62 and 67 µg m -3 , respectively, for the two time periods.

Long-range transport as simulated by the regional model
The impact of sources outside the Curitiba municipality, including both nearby industrial sources in the Araucária 10 municipality southwest of Curitiba, as well as remote contributions from the state of Paraná and the rest of the continent, was simulated by the regional model BRAMS 5.2 (see Section 2.5). Anthropogenic emissions, including the nearby industries, largely contributed to PM 2.5 levels, while biomass burning presented a small impact (Table 5). Since the IAP inventory did not include BC emissions, thesey were taken from the coarse resolution global EDGAR-HTAP database. Consequently the BC contribution from the industrial emissions just outside Curitiba is likely underestimated very poorly described by the 15 model.

Simulations of local traffic impact inside the street canyon
We ran the street canyon dispersion model OSPM separately for public transport and private vehicles for the period from July 25 to August 24, 2016, with output at the MD site. Table 6 shows the comparison of measured and simulated NO x , PM 2.5 and BC impact of local traffic inside the street canyon, using emission factors from the literature (Tables 2 and 3). For 20 NO x , there was a relatively good agreement, but for both PM 2.5 and BC the simulated impact wasconcentrations were much lower than what the measurements indicated. Since uncertainties related to the meteorological data and model assumptions are equal for all three compounds, it is reasonable to assume that the larger uncertainties found for BC and PM 2.5 are caused by uncertainties in the effective emission factors. In addition, for PM 2.5 , there is a fraction related to road, tire and break wear, which should be added to the exhaust emission factors used in the model simulation. To some degree the 25 underestimated effective emission factors for the private vehicles may be related to fleet composition (for the public transport the fleet composition was known in detail). However, heavy duty vehicles that do not form part of the public transport are have restricted circulation in the studied area, thus reducing this source of uncertainty.
Since BC measurements were available as hourly data, it was possible to perform a multiple regression analysis with measured BC increment as dependent variable and the two simulated hourly time series of the contributions from buses and 30 private traffic as independent variables. The regression analysis suggests that the public transport signal (bus impact) should be multiplied by a factor of 1.2 and the private traffic by a factor of 5. The regression, with a forced demanding zero intercept, was based on 696 hourly data points from the monitored increment between roof and street stations, and from model output for the two simulations, yielding an adjusted R 2 of 0.78. Fig. 87 and the mean values of Table 7 show that with these corrected emission factors, the model output matches the measured street increment in the street canyon. With the assumed fleet composition, the effective BC emission factor for the mixed private vehicle fleet was corrected from 4 to 19 mg veh -1 km -1 . 5 There were no hourly PM 2.5 data to perform a similar regression analysis.; hHowever, scaling up the summed model output of the bus and private vehicle contributions to PM 2.5 by a factor of 5 yielded a local contribution similar to the monitored street canyon increment.

Urban and regional simulations of PM 2.5 and BC
The Gaussian urban model was applied with the corrected emission factors for the on-road vehicles, and the results were 10 subsequently added to the regional LRT contributions. Table 8 shows the contributions at the two sites and Fig. 9 displays the mean spatial distribution of the simulated PM 2.5 and BC concentrations. The emission inventory show local PM 2.5 emissions from the transport sector (public and private together) within the urban model domain over Curitiba of 643 tons year -1 , which can be compared to 329 393 tons year -1 from the smaller industries located inside the city and 1,600 tons year -1 (calculated as 70% of officially reported PM 10 emissions) from the industrial Araucária area just southwest of the city. For 15 BC, we estimated 18 tons year -1 from public transport and 375 tons year -1 from private traffic (no information was available on the industrial contribution).

Pollution levels in Curitiba in general
The information on PM 10 and NO 2 levels from the official monitoring network in Curitiba revealed mean annual PM 10 20 concentrations between 15 and 30 µg m -3 , which are between the values presented in the WHO global assessment for highincome countries and for low-to middle-income countries in the Americas (see Fig. 3  The monitoring campaign was only a month long and planned to be representative of wintertime conditions in the southern 25 hemisphere, when the highest pollution levels are usually observed. However, the meteorological conditions, and likely also a smaller contribution from long-range transport than usual, caused the PM 10 and NO 2 levels to be considerably lower during the 2016 field campaign, as compared to average concentrations in previous years. This indicates that monitored PM 2.5 and BC concentrations reported for the 2016 winter month campaign could also have been lower than what is normally the case for the months of July and August, likely more comparable to annual mean values of full years as given by earlier studies. 30 The urban background PM 2.5 concentrations registered at the roof top in the city center (7.3 µg m -3 ) agreed reasonably well with the 10.3 µg m -3 reported from a year-long measurement in 2014-2015 at the Federal University campus (Polezer et al., 2018), considering the fact that the station was located approximately 100 m from an interstate highway. Measurements

High PM 2.5 levels in the residential area (station SC)
At the residential station SC, the mean PM 2.5 concentration (25.3 µg m -3 ) was significantly higher than in the urban background of the city center (7.3 µg m -3 ). The existence of an additional local source of PM 2.5 , not included in the emission inventory, can be deduced from the mapped PM 2.5 concentrations of Fig. 89 (left), which does not indicate any levels in the range of 20-30 µg m -3 around the SC station. The mean OC/EC ratio of 4.4 (range 3.3-9.0) found in PM 2.5 sampled at SC 10 indicates that the dominant local emissions in this area were unlikely to originate from diesel traffic emissions along the nearby ring road, since the EC contribution for diesel vehicles can be expected to be larger than the OC part (Harrison and Yin, 2008). Comparatively, a monitoring campaign performed at four urban sites in São Paulo (Monteiro dos Santos et al., 2016) reported OC levels between 2.65 and 3.437 µg m -3 and EC levels between 6.11 and 1.50 µg m -3 , yielding OC/EC ratios of 1.57, 1.88, 1.89 and 0.56. The lowest OC/EC was observed within a street canyon, suggesting a large contribution 15 from diesel-fueled vehicles. Thus, the higher OC/EC ratio found at the Curitiba residential station, makes it unlikely that traffic emissions and the closeness to the ring road were the cause to the higher PM 2.5 levels. Instead, the high OC levels may indicate the presence of wood burning around the SC station. Curitiba municipality lacks information on such residential biofuel devices, but the staff operating the monitoring station occasionally reported the smell of wood smoke and noted the presence of several houses with small chimneys in the neighborhood. 20

BC levels in Curitiba
The simulated BC levels in the urban background, i.e., excluding street canyons and the immediate vicinity of the major roads, were fairly homogenous over the city (Fig. 98, right). Also the few existing measurement data points indicated a spatial homogeneity, with mean values just above 2.0 µg m -3 at the sites MD roof and SC reported in this paper and also at 25 the university campus, considering the year-long dataset from 2014-2015 (Polezer et al., 2018). Within street canyons, the BC levels were locally raised Krecl et al., 2016) and also close to highways with diesel traffic . Even if BC levels show to be spatially homogenous in the urban background, there were large temporal variations due to meteorological conditions, changes in the long-range contribution and likely due to some unidentified BC sources inside or just outside the city. The temporal evolution of BC levels at station MD and SC in Fig. 65  30 shows high BC levels at both stations during the second and third weekends of the monitoring campaign, however, considerably higher BC peaks were observed at the residential site SC. These enhanced BC peaks at the SC station may have resulted from wood combustion in the vicinity of the SC station, even if the BC levels were not increased as much as the OC and PM 2.5 levels.
The mean BC concentration of 5.5 µg m -3 observed at MD street level was similar to what was registered during weekdays in a Stockholm street canyon in 2006 (5.1 µg m -3 ). A later comparison in 2013 for the same site in Stockholm showed reduced levels (2.2 µg m -3 ) explained by technologically improved vehicles trafficking this street canyon (Krecl et al., 2017). This 5 gives an indication of what is possible to achieve with regulations on vehicle technology.

Long-range contributions as assessed by the regional model
The regional model was used to estimate the long-range contributions of PM 2.5 and BC to the city. The emission inventory of industrial sources in Paraná only included PM 10 emissions, which required motivated thean assumption that 70% of those 10 were in the fine particle size range. Table 5 thus gives the magnitude of the PM 2.5 impact from industrial sources and also the PM 10 impact, which can be seen as an upper limit if all industrial PM emissions are found in the PM 2.5 fraction. The regional model also indicated a rather small (0.5 µg m -3 ) contribution of biomass burning to PM 2.5 in Curitiba for the monitoring campaign period. Unfortunately, the regional model did not have access to complete BC emission data, since the Paraná inventory of industrial emissions only covered PM 10 . With only the large-scale global EDGAR-HTAP emissions, the BC and 15 OC content in the long-range transported air arriving to Curitiba was probably clearly underestimated by the model (see Section 4.7 for the discussion of measured versus simulated BC levels in the urban background of Curitiba).

Contribution to BC levels from local traffic
An important aim with the ParCur assessment in Curitiba was to determine the emission of PM 2.5 and BC from sources 20 inside the city. For that purpose, the street canyon measurements were used to evaluate in situ the emission characteristics of the vehicle fleet in Curitiba. With emission factors taken from the literature as input to the dispersion model, there was good agreement between the model outputs and the measured NOx concentrations, but the BC increment within the street canyon was underestimated by a factor of 3 (Table 6). The OSPM model has been used and evaluated in many urban environments and an extensive comparison showed good results for NO x (Ketzel et al., 2012). If assumed that the model gives reasonable 25 results, then the good resemblance between simulated and monitored NO x values indicates that the NO x emission factors should be fairly accurate, but also that the BC emission factors are erroneous. With this as motivation, new emission factors for BC were determined through regression analysis. An interesting fact is that the regression analysis confirmed the magnitude of the BC emission factors of the buses, whose fleet composition, age and equivalent Euro class we know in detail. This is a result that can be used in other Brazilian cities where the characteristics of the public bus fleet is equally 30 known in its details. However, the regression pointed out a large underestimation (factor of 5) of the mixed private vehicle fleet emissions, for which the composition and technical status was much less known. A factor of 5 of difference in an emission factor may be seen as a too large discrepancy. However, BC emission factors show a large scatter and both the original 4 mg veh -1 km -1 and the corrected 19 mg veh -1 km -1 for the mixed private vehicle fleet circulating through the MD street canyon fit into "real-world" emission factors reported in the literature. Through chasing individual vehicles, BC emission factors spanned from 10 to 32 mg BC km -1 for gasoline cars (Ježek et al., 2015), while for heavy-duty diesel vehicles they went from 53 mg BC km -1 (Park et al., 2011) to values 10-fold higher ). An assessment of 5 real-world BC emission factors performed in a street canyon in Stockholm (Krecl et al., 2017) for the year 2006 (when BC levels within the street canyon were similar to those reported for M. Deodoro in Curitiba) found a BC emission factor for gasoline cars of 11 mg veh -1 km -1 , which is significantly higher than the 0.15 mg veh -1 km -1 used originally in the present study (Table 3) and can therefore justify the higher emission factor coming out of the regression. A lesson learnt for other Brazilian cities is thus the necessity to assess local BC emission factors for the private vehicle fleet, using similar 10 experiments as in Curitiba.
The boxplot comparison between mobile measurements of BC along the four biking routestrajectories in the city center and the measured BC concentrations levels at the fixed MD street canyon station revealed the large heterogeneity in traffic rates and street layouts in the city center when compared to the a fixed site in a busy canyon street. The variability in traffic rates create differences in emission rates at microscale level (individual street canyons), and varying building heights will also 15 influence the immediate dilution of air pollutants emitted by local traffic.

Contributions to PM2.5 levels from local traffic
For PM 2.5 , which at the MD station was only measured as daily averages, it was only possible to compare the simulated total impact of all vehicles in the street canyon with the measured increment (i.e., the difference between the more polluted air in 20 the bottom of the street canyon and the cleaner atmosphere at the rooftop). Also here the simulated PM 2.5 contribution of the local traffic was largely underestimated when compared to the measured increment. For PM 2.5 , contributions from nonexhaust particles generated by the vehicles are also expected, namely those consisting of wear particles from brakes and tires, together with wear of the street pavement itself, which were not included in the emission factors of Tables 2 and 3. However, since the aim was to estimate the total PM 2.5 emissions, no separation was made between exhaust and non-exhaust 25 PM 2.5 originating from the traffic in Curitiba. Since it is not possible to perform a regression that can apportion the impact on PM 2.5 of public transport and private vehicles (lack of hourly data), a general correction with a factor of 5 was applied to both buses and the mixed fleet of private vehicles.
The boxplot comparison between mobile measurements of BC along the four biking trajectories in the city center and the measured BC levels at the fixed MD street station revealed the large heterogeneity in traffic rates and street layouts in the 30 city center when compared to a fixed site in a busy canyon street. The variability in traffic rates create differences in emission rates at microscale level (individual street canyons), and varying building heights will also influence the immediate dilution of air pollutants emitted by local traffic.

Spatial distribution of PM 2.5 and BC over Curitiba (comparison between simulated and measured concentration levels)
A goal of the ParCur project was to map the annual concentrations of PM 2.5 and BC over the city, identifying the contribution of local sources and the magnitude of the long-range transport. As indicated by Table 8, a complete picture has  5 not been possible to achieve. For PM 2.5 , the summed model output attained 5.1 µg m -3 compared to 7.3 µg m -3 measured in the city center. The missing 30% contribution could possibly be attributed to sources not included in the emission inventory (e.g. restaurants and residential sources), and also to underestimated particle mass in the regional model output (which did not include secondary aerosol formation). At the residential site, the simulated PM 2.5 levels only covered 20-25% of the measured concentrations, indicating a strong local source not being included. As discussed earlier, there are indications of 10 wood combustion taking place in this residential area on the city's outskirts, but not in its center.
For BC, the simulated concentrations constituted 56% of the measured concentrations in the city center and exactly half at the residential location. Most of this difference is likely due to a too low BC concentration in the simulated long-range contribution, since the offical emission inventory for Paraná lacked industrial emissions of BC (e.g., those in Araucária just southwest of Curitiba) and the emission input had to be taken from the coarse global database EDGAR-HTAP. For the 15 residential site, the existence of residential wood combustion would also contribute to increased BC levels. However, not as large as for PM 2.5 , this since high OC/EC ratios are expected in fumes coming from wood combustion.
The few monitoring points and the short sampling period contributed to certain uncertainties in our conclusions concerning the PM 2.5 and BC emissions and concentrations in Curitiba. For the city's inner core, the information from the fixed MD station and the mobile monitoring, supported by the model results, should together allow a good description of emissions and 20 concentrations. However, the higher air PM 2.5 and BC concentration indicated in the model output (Fig. 89) along the ring road "Contorno" and around the industrial area southwest of Curitiba, could not be verified by local monitoring and also included various methodological simplifications, such as the vehicle fleet composition, emissions from the industries, that makes the concentrations indicated for this area more uncertain. In connection to this problem, great uncertainty subsists in knowing the actual characteristics of the gross polluters, notably old trucks (Fig. 109) in transit through Curitiba's ring road. 25 Many of them are not registered in the state of Paraná and, consequently, not taken into account in our inventory.

Lessons learnt from the Curitiba case study and its applicability for other Brazilian cities
The outcome of the monitoring campaign was positive for the BC part, whereas the PM 2.5 sampling at the central street canyon station implied two difficulties. The major one was the technical problems that impeded the calculation of the 30 increment from roof to street and the necessity of using BC data as a proxy for PM 2.5 . Also, the fact that PM 2.5 data had daily resolution made it difficult to separate public transport and private vehicles, which had different temporal variation during the day. Since the determination of local emissions and their impact rely heavily on monitored data, it is fair to conclude that the BC results, both monitored and simulated, should be seen as more robust than the PM 2.5 results.
This study highlights the need to develop emission inventories for PM 2.5 and BC at local, regional and national levels that are currently missing in Brazil. Other local activities such as small combustion from backyard burning and wood burning appliances, and emissions from pizzerias, bakeries and steakhouses should be also considered in the inventories. 5 Data on airborne fine particle emissions and concentrations in cities are valuable to traffic and air quality managers, urban planners and landscape architects, health practitioners, researchers, and ultimately to legislators and decision makers. Most Brazilian cities lack this kind of information which hinders an accurate assessment of their ambient concentrations, the source apportionment and the potential health outcomes. Air quality monitoring networks with sufficient spatial coverage are unlikely to be available in Brazil in the near future. Hence, methods, instruments and models to identify air pollution 10 hotspots and their sources are an urgent need to safeguard the population exposure to toxic airborne species. This paper, as a result of the Brazil/Sweden cooperation materialized in the ParCur project, addressed this gap taking Curitiba as case study.
The integrated approach used in the study can be implemented in other Brazilian cities as long as an open data policy and a close cooperation between municipal authorities and academia can be achieved. • Local emission factors for BC were determined for on-road vehicles.
• BC emissions from local traffic contributed to about half of total levels in Curitiba, the restremainder being predominantly caused by sources outside the Curitiba municipality. Typical urban background levels of BC concentrations were found around 2.0 µg m -3 , raising to > 5.0 µg m -3 close in to central street canyons. 20 • Urban background PM 2.5 concentrationslevels in central Curitiba were 7.0-8.0 µg m -3 , doubling inside street canyons. In a residential area, concentrations three-to 3-four-fold4 times higher levels were observed, most likely due to local biomass or waste combustion.

Limitations
Some important limitations of this PM 2.5 and BC assessment in Curitiba have been identified: 25 • Measurement campaign limited in time (one month) and space (two fixed stations and mobile measurements along four mobileroutes covering an area of 4.8 km²)measurements on bike): An attempt was made to relate the one month-long campaign data to the wintertime conditions observed during the three preceedingpreceding years, this in order to assess the representativity of our collected PM 2.5 and BC data. The two fixed sites were well selected, however there is a need for future monitoring campaigns atin additional locations (e.g., in the more polluted areas • Technical failures during the campaign contributed to increased uncertainties in the PM 2.5 assessment in the street canyon. Also here a future campaign could give more reliable information on emission factors for the Curitiba vehicle fleet. • While the model input information for simulating the impact of public transport was very detailed and of high quality, there was a lack of information on fleet composition for the private vehicles circulating the road network. 5 • Lack of emission information ofon some local sources of potential importance (e.g. restaurants, residential wood combustion), as well as regional information on PM 2.5 and BC emissions from the industrial sector.

Funding sources
This study has been supported on the Swedish side by the Swedish Ministry of Environment and Energy, and from the    Fig. 23: Domains for the regional model (left) and the urban dispersion model (right). For the latter, building heights above 10 m are marked (right). Over the remaining built-up city areas the building height was set to 6 m.        (10 m Table 3. Emission factors for private vehicles were taken from the report by the European Environment Agency (EEA, 2016), where BC emission factors are expressed as percentages of PM 2.5 emissions (see Table 3 -91 in EEA, 2016  August 76% * At BOQ only PM 10 was measured and at STC only NO 2 . 5 Table 5. Contribution of long-range transported pollution (LRT) to MD in the city center and at the residential site SC. Period: July 25 th to August 24 th , 2016.  2.3)).

MD rooftop SC Emission sources
5 Table 6. Comparisons between simulated and measured air pollutants from local traffic inside MD street canyon. Note that the measurements are reported here as street increments (given as the difference between street and roof level concentrations) to be comparable with the model output, which only gives the impact of the road traffic inside the street canyon. 5  Table 7. Measured street increments and simulations of local traffic impact inside the canyon where the monitor station MD (street level) was located, after correcting the emissions factors.

Model simulation Period
(µg m -3 ) (µg m -3 ) Buses private Total 25 th July -24 th August, 2016 PM 2.5 9.3 8.0 * 25 th July -24 th August, 2016 BC 3.2 0.8 ** 2.2 ** 3.0 ** * Total simulated PM2.5 output of Table 6 corrected with a factor of 5. 5 ** Bus impact of BC corrected with a factor of 1.2 and private traffic impact of BC corrected by a factor of 5.0, with values given by a regression analysis of hourly data.