Local and Regional Contributions to Black Carbon Aerosols in a Mid-Sized City in Southern Brazil

Black carbon (BC) concentrations were monitored at three sites (suburban, street canyon and urban rooftop) in a midsized Brazilian city, from August 2014 to January 2015. The suburban site presented weak diurnal cycles, suggesting little influence of motorized traffic, but distinctively large 95 percentile concentrations, reaching values as large as 11.04 and 3.34 μg m in the evenings of the dry and wet season (respectively), likely attributable to local waste burning. Moreover, higher BC concentrations at the suburban site were observed throughout the dry period, primarily caused by long-range transport (LRT) of smoke from the central part of Brazil and neighboring countries, carried by WNW, SSW and NNE winds. Local traffic was by far the most important source of BC in street canyon, with mean hourly peaks of 5.8 μg m (at 7:00) and 4.6 μg m (18:00), coinciding with rush hour periods. The rooftop data showed a mean peak of 1.4 μg m at 7:00, reflective of traffic on a busy avenue adjacent to the site. Meteorological data clustered into groups of similar air temperature (Tair) and relative humidity (RH) showed that BC concentrations were highest (18.3 μg m) at the suburban site during the evenings of dry (RH ca. 20%) and hot days (maximum Tair ca. 30°C). Diurnal concentrations in the canyon and rooftop were linked to traffic patterns and showed no clear linkage to meteorological conditions. This study shows that the BC concentrations in the city are highly variable and that air quality diminishes considerably due to sporadic waste burning and LRT of biomass smoke, even in neighborhoods with little motor traffic. While air pollution due to transboundary smoke is more difficult to abate, these results suggest that targeting local backyard burning and traffic volume would lead to a depletion of BC concentrations in the city.


INTRODUCTION
Today, air pollution concerns affect not only large cities and mega-metropolises (populations exceeding 10 million) but also mid-sized cities, as urban populations within this category have increased considerably over recent years.For example, Brazilian mid-sized cities -with populations ranging from 100 to 600 thousand-grew between 2000 and 2007, while small and large cities showed a decline in population (IPEA, 2008).According to the United Nations (1997), the growth of urbanization in Latin America has been accompanied by the proliferation of slums and squatter settlements on the outskirts of the urban core, forcing people to drive a greater number of vehicles, over longer distances.The consequences of air quality deterioration for the urban population is more acute in developing countries with a tendency of operating old and badly-maintained fleets.All in all, motor vehicle traffic is the most important source of air pollution in cities, and according to Lamarque et al. (2010), transportation is the predominant source of anthropogenic black carbon (BC) in Latin America.Johansson et al. (2007) also showed that road traffic emissions are a major contributor to the number concentration of ultrafine particles in urban environments, because the bulk proportion of particles generated from gasoline and diesel combustion are smaller than 60 nm and 100 nm, respectively (Kittelson, 1998;Gouriou et al., 2004).Diesel-powered vehicles, which are usually a smaller fraction of the fleet, make a proportionally large contribution to total number concentrations (Kittelson et al., 2006).For example, Krecl et al. (2015) showed that 73% of BC particles in Stockholm come from diesel emissions, especially in the early hours at weekends.
A complicating factor in accurately assessing the concentration of urban air pollutants is that traffic is concentrated in the city's inner core, which can comprise a complex configuration of canyons, avenues and parks.The combination of designs, materials and geographical attributes impact the air flow within the urban canopy, affecting the concentration of pollutants in this layer.Longley et al. (2003) and Krecl et al. (2015) showed that rooftop winds blowing perpendicular to the street or with directions to the street axis exceeding 30°, trigger a recirculation vortex in which the air aloft flows into the canyon windward sector and across the street, causing higher concentrations of air pollutants within the leeward side.Moreover, different traffic volumes, types of fuels and fleet ages contribute to an interwoven diurnal pattern of air pollutants.
BC particulate is one of the most harmful air pollutants found in urban environments and is emitted by the incomplete combustion of carbon-containing materials such as biomass and fossil fuels.BC is made up of chains of spherules, has diameters in the 10-50 nm range, has a non-zero imaginary part of the refractive index, absorbs visible radiation, has a vaporization temperature near 4000 K and is insoluble in water and common organic solvents (Bond et al., 2013).In the early 1980s the World Health Organization (WHO) recognized the effects of BC particles on human health and formulated the first guidelines for exposure limits to BC.Because of its small size, BC particulate can reach the lower respiratory tract, potentially carrying toxic chemical species deposited onto their porous surface.Diesel fumesknown for containing a high load of BC-have been reclassified by the WHO from 'probably carcinogenic' to 'carcinogenic'.Despite their short life span of 1 to 2 weeks (in the absence of precipitation), BC absorbs incoming solar radiation, and is the second most important man-made agent of climate change, following carbon dioxide (Bond et al., 2013).
In addition to BC particles emitted by industries, power plants (ALA, 2011) and traffic, urban areas can be subject to plumes of BC from local open burning and the outflow of smoke from biomass fires (e.g., Saarikoski et al., 2007;Targino et al., 2013).Open burning involves the combustion of any matter in open space and the emission of pollutants (BC, organic carbon, inorganic species, polyclinic aromatic hydrocarbons, dioxins, among others) directly into the ambient air without passing through any duct filtration, causing nuisance of smoke and odor.
The main burning activities that affect the urban air quality are combustion of crop residue, land clearing debris and house waste (Lemieux et al., 2004).In Brazil, preharvest burning of sugarcane straw affects the air quality of cities both near and downwind of the burning areas (Artaxo et al., 1994;Reinhardt et al., 2001).Burning off of agricultural debris and domestic yard waste (practiced in barrels, open pits or open piles) is common in developing countries.In Brazil, this practice is unlawful, but loosely controlled, and frequently observed in rural areas and neighborhoods with poor waste handling logistics or to avoid paying for waste collection service.Domestic waste comprises a mix of organic (e.g., food and garden debris) and non-organic waste (e.g., plasticized wrappings and other polymeric materials) which burn with uncharacterized emission factors.While governments are engaged to curb industrial and traffic emissions in response to environmental regulations, BC emissions from open burning dominate global emission inventories, especially between 20°N and 30°S (EPA, 2012).
BC measurements are scarce in Brazil and, to the best of the authors' knowledge, it is not monitored continuously at air quality stations.Most information on BC concentrations comes from a few dedicated studies conducted in the Amazon (Martins et al., 1998) and in large cities (de Miranda et al., 2010;Backman et al., 2012).Nevertheless, Brazil ranks fourth in the world as a BC emitter and the main sectors are biomass burning (61%) and motorized transportation (19%), according to global emission inventories (EPA, 2012).However, a large uncertainty is associated with these estimations due to inaccurate information on BC emission factors, fuel consumption, type of combustion devices, features of vehicles and engines, and the coarse spatial resolution of existing global aerosol models (Bond et al. 2004;Ramanathan and Carmichael, 2008).
We conducted this BC monitoring study in Londrina -a mid-sized city in southern Brazil-where the following sources of BC co-exist: (i) The vehicle fleet amounts to approximately 362,000 vehicles (As of June 2015), with 52% gasoline-powered units, 33% of flex engines (that run on gasoline, alcohol or any blend of these fuels), and a diesel share of 7%, comprised mostly of heavy-duty vehicles.In Brazil, pure gasoline is blended with 25 ± 1% (v/v) anhydrous ethanol and is called gasohol (Ré-Poppi et al., 2009).All diesel used for on-road transport has a blend of 6% of biodiesel and, since 2013, low Sulphur (S) diesel S10 (S content of 10 mg kg -1 ) is available along with S500 (S content of 500 mg kg -1 ).(ii) Fire foci are frequently observed due to domestic waste burning, clearing of leaves and prunings and disposing of crop residues in small farms located near the city perimeter.In 2013, the Environment Secretariat (SEMA) registered 280 complaints of illegal open fires.(iii) The city is also subject to long-range transport (LRT) smoke linked to the annual cycle of pre-harvest burning of sugar cane straw and savannah (cerrado) in neighboring states, typically from August to October.The core objective of this study was to analyze the spatioseasonal patterns of BC concentrations in different urban microenvironments affected by both local and regional pollution and their linkages to meteorological variables.This characterization is well-aligned with the recommendations made by the General Assembly of the World Medical Association (WMA) held in Durban (South Africa) in October 2014, which includes: a) monitoring and limiting the concentrations of BC in urban areas, b) increasing public awareness on the health damage caused by BC particle emitted by diesel vehicles, and c) developing strategies to reduce people's exposure to BC in aircraft passenger cabins, trains and homes (WMA, 2014).

Study AREA
Londrina is a city with 548,000 inhabitants located in the state of Paraná, southern Brazil (lat.23°19′S, long.51°08′W, alt.630 m).It was founded in 1934 and despite being a young city, Londrina faces environmental problems comparable to older and larger cities in the country.The motorized fleet grew 85% in the last decade, and is dominated by passenger vehicles (78.3%) that circulate with an average of 1.47 persons per vehicle.Motorcycles increased 67.6% in the period 2005-2014 as an affordable alternative to the inefficient public transportation system.The motorization rate reached 661 vehicles per 1000 inhabitants in June 2015, which is much higher than the national rate (436 vehicles per 1000 inhabitants).Londrina has a humid subtropical climate (Cfa in the Köppen-Geiger classification) with an annual mean temperature of 21.0°C and annual mean precipitation of 1,630 mm.Rainfall occurs throughout the year, but mostly in the summer (December to February) whereas winters are drier.In winter, the average air temperature (T air ) and relative humidity (RH) range from 11.6 to 25.8°C, and 62 to 75%, respectively, and from 19.0 to 29.7°C, and 72 to 77% in summer.Insolation is greatest during winter time (mean value of 225 h) when cloud coverage is lowest (Targino et al., 2014).

Sites Locations
We explore the behavior of BC concentrations using aethalometry data collected from August 2014 to January 2015 at three sites subject to both local and regional pollution sources.The measurements were conducted in the dry and wet months to capture different patterns of BC concentrations governed not only by sources typically found in urban environments (e.g., traffic emissions), but also by sporadic (e.g., open backyard burning) and seasonal (LRT) emissions.A description of the measurement sites follows and their geographical location is displayed in Fig. 1.

Federal University of Technology (UTF) Campus
The UTF campus is located ca. 8 km from Londrina's city center, on the eastern edge of the city, with the closest neighborhood consisting of sparsely built, detached houses about 400 m away, across wheat and soybean fields.The major motor vehicle pollution source comes from a paved road which gives access to the campus and to a residential cluster development, with a traffic volume of 2,700 vehicles day -1 on weekdays (8.4% are diesel heavy-duty vehicles).The measurement site is located 165 m from this road in straight line.Visual inspection indicates that open fires occur frequently near the campus during the dry season when residents burn branches, leaves, domestic debris and also plastic from copper wiring.

Higienópolis Avenue
This is a 2.8-km arterial road aligned in the north-south direction and collects traffic from many small streets, crossing areas of commercial buildings and high-rise multifamily units.The road has two lanes in each direction and lies on terrain with varying heights (maximum elevation difference of 100 m).The daily traffic volume amounts to ~25,000 vehicles day -1 on weekdays (2.4% are diesel heavy-duty vehicles) and up to 3,700 vehicles h -1 in the evening rush hours.

Sergipe Street
Sergipe is a heavily-trafficked one-way street aligned in the east-west direction, located in the city center, with intense commercial activity from Monday to Saturday.Measurements took place within a flat transect that presents a canyon structure (width of 14.5 m and length of 113 m), with buildings continuously flanking both sides of the street with heights between 3.9 and 9.7 m.The traffic is organized in two lanes with a speed limit of 40 km h -1 , and controlled by traffic lights situated at both ends of the transect.Thus, start-stop driving is the general driving pattern and congestion is frequently observed during rush hours.Manual counting conducted at specific times revealed a traffic rate of 1,050 vehicles h -1 in the afternoon, with the following share: passenger cars and light-duty vehicles (66.6%), motorcycles (24.0%), buses (7.8%) and trucks (1.6%).

Instrumentation
BC concentrations were measured using two instruments: a benchtop 7-wavelength (λ) aethalometer model AE42 (Magee Scientific, Berkeley, CA, USA) and a 1-wavelength microaethalometer model AE51 (AethLabs, San Francisco, CA, USA), operated as described in Table 1.The bench top aethalometer has been in use since the 1980s and has been cited in over 400 peer-reviewed articles (as of September 2015 on Web of Science).A thorough description of its principle of operation can be found elsewhere (e.g., Hansen et al., 1984).In short, particle-laden air is drawn at a constant volumetric flow rate (Q) from the environment via a tube inlet and the particles accumulate on a filter spot of area A. The particles on the filter attenuate the incoming radiation beam (I 0 ) and the attenuation ATN is calculated as: where I is the intensity of the radiation detected after passing through the filter.
The instantaneous wavelength-dependent aerosol absorption coefficient (b abs ) is calculated using the change in light attenuation ∆ATN in a time interval ∆t: The method assumes that ATN is linearly proportional to the BC loading on the filter and that there is no other absorbing material on the sample than BC.The BC concentration is related to ATN via the wavelength-dependent cross sectional absorption coefficients σ λ [m 2 g -1 ]: (3) The model AE42 uses a 7-wavelength radiation source (370, 470, 525, 590, 660, 880 and 950 nm) which illuminates a teflon-coated quartz filter tape (Pallflex Q250F).The model AE51 is a portable unit which operates with an 880-nm radiation source and uses a filter ticket made of T60 teflon coated borosilicate glass fiber material.The relationship in Eq. ( 3) is valid for low attenuation ranges and becomes non-linear as attenuation increases, which is attributed to the filter loading effect, that is, the loss of sensitivity of the measurement with the loading of the filter spot (Gundel et al., 1984).This effect is also known as the shadowing effect since the existing particles may shadow those freshly-collected which, in turn, are not exposed to the same intensity of light, leading to an underestimation of the measured signals.
Another artefact in filter-based BC measurements is the attenuation of I 0 due to multiple light scattering when the filter is relatively unloaded (Weingartner et al., 2003), leading to an enhancement of the optical path and, thus, to erroneously enhanced values of light absorption.Many studies perform a post-processing to correct the scattering effect by using data of collocated aerosol scattering coefficient measurements (e.g., Petzold et al., 2004;Arnott et al., 2005), while a compensation for the loading effect is applied in terms of a correction (called k factor) obtained by using the raw BC dataset (Weingartner et al., 2003;Virkkula et al., 2007).The k factor depends not only on the season and the site but also on the immediate local sources, which also change along the day, why it is recommended to extract k factors from the raw BC data.We only performed the correction for the loading effect, in which the wavelengthdependent corrected BC concentration is expressed as: where BC 0 (λ) is the (non-corrected) concentration given by the aethalometer.For each filter spot a k factor is calculated as:  (Virkkula et al., 2007).When an attenuation threshold is reached (ATN max ), the filter tape advances automatically in model AE42 (exposing a new clean filter spot) whereas the filter ticket has to be changed manually in model AE51.ATN max was set to 75 for AE42, except for some days during UTF dry when ATN max reached 125, hence these days were assessed separately.We replaced the AE51 filter tickets at every visit to the site, and the typical ATN max was 60.
Because ATN is not exactly zero even at the first data point of the new spot (unless BC concentration equals zero), in practice the k factor was calculated using data of the last 10 minutes of filter spot i and data of the first 10 minutes of filter spot i+1.Eq. ( 5) may fail to yield realistic k values when the concentrations changes rapidly or when the concentration is higher immediately before the spot change than after, in which case k becomes negative.Thus, we used averaged k values for each instrument, site and sampling period (Table 2), which yielded more realistic corrections than individual k values.Most results of this study focus on BC concentrations at the coincident wavelength λ = 880 nm of both instruments, while BC concentrations at λ = 370 nm are shown for the suburban and urban sites only.Since the AE51 had no size-selective inlet and the AE42 operated with a 2.5-µm cyclone, we conducted a collocation comparison in ambient air during three days.The one-hour average data were found to be highly correlated (R 2 = 0.896), which indicates that most BC mass is in the fine mode and the use of a 2.5-µm cyclone has little effect on the comparison of the datasets reported in this study.Meteorological variables (T air , RH, wind speed WS, wind direction WD, and precipitation) were measured at UTF campus.All times are expressed in local time (UTC-3) and as 24 hour format.

General Overview
The average air temperature and rainfall of August and September 2014 were above the 1976-2014 mean climatological values furnished by the Agronomic Institute of Paraná (IAPAR), with anomalies of 1.3°C and 62.6 mm.September marks the beginning of the rainy season and the positive rainfall anomaly is due mainly to the events that occurred towards the end of the month.November 2014 was unusually dry (35 mm below average), with sporadic rainfall that occurred as heavy isolated precipitation events.December 2014 and January 2015 received ca. 25 mm less rainfall and were 1.3°C warmer than average.Rainfall was well-distributed during this period, apart from an intense episode on 23 December 2014 that accumulated 88.4 mm.Table 3 shows a summary of the meteorological conditions during the sampling periods.

Diurnal Variation of BC Concentrations
The BC data show a large intra-site variability, with the highest mean and median BC concentrations in the street canyon and UTF dry and lowest values at the rooftop and UTF wet (Table 4).The street canyon and UTF dry also show the largest standard deviation (SD) which, in the case of the former, is due to the large variations in traffic volume and emissions, while the latter is due to the sporadic nature of the sources close to the university, alternating between clean and dirty spells during the day.The intra-site and seasonal variability of BC concentrations is evident in the mean BC diurnal cycle (Fig. 2(a)).The canyon data show a well-defined pattern matching the traffic rush hours, peaking at 07:00 (5.84 µg m -3 ) and 18:00 (4.55 µg m -3 ).This street section gives access to the city's bus station and has a large volume of buses throughout the day, especially in the rush   hours when the volume increases to serve the terminal's commuters.The rooftop measurements show a faint mean diurnal cycle with a dome between 7:00 and 11:00, and a peak of 1.47 µg m -3 at 07:00, matching the morning rush hour.The effects of the dry and wet seasons are evident at the university campus, where the traffic contribution is rather low and the BC concentrations are governed by backyard burning and LRT.A hallmark in both series is the peak at 20:00, reaching mean BC concentrations of 2.39 µg m -3 in the dry season and 1.18 µg m -3 in the wet season.This feature and the causes behind it are discussed further on in this section.
Because BC concentrations are highly variable in time, the mean hourly diurnal cycle may mask patterns related to processes inherent to certain periods of the day, such as sporadic burning and increase in traffic volume.To unveil features, we have split the datasets into weekdays (Monday to Friday) and weekend (Saturday, Sunday and holidays) and also calculated the 95 th percentile (P95) of BC daily cycle (Fig. 2(b)).In spite of representing only 5% of the dataset, P95 is an indicator of extreme values and useful to estimate human exposure and pinpoint periods when people will be more subject to extreme BC concentrations.On weekdays, P95 peaks at 07:00 and 18:00 (15.83 and 11.89 µg m -3 , respectively) at the canyon site.The shape of the weekend data resembles the weekday data, except for the broader and lower peaks in the morning and afternoon rush hours.On Saturday mornings most businesses are open until 13:00, and traffic volume is high.The P95 peak is shifted to about 10:00 consistent with the local practice of people going to the city center towards mid-morning.On Saturday afternoons only a few shops are open in the area and BC data seems to respond to the more sporadic nature of motor traffic.The rooftop weekday data show a P95 peak of 4.35 µg m -3 between 07:00 and 10:00, and secondary peaks between 20:00 and 02:00, probably associated with late evening traffic.An interesting feature in UTF dry is the P95 shoulder between 18:00 and 21:00 with a peak of 11.04 µg m -3 on weekdays and 3.87 µg m -3 at weekends.Visual inspection by the university community reported smoke in the vicinity of the campus, and the authors approached some fire foci and found out that residents also burn off the plastic insulation of copper wires.This practice is reduced in the rainy season and the P95 peak is much lower in UTF wet (3.34 µg m -3 on weekdays and 2.89 µg m -3 at weekends) (Fig. 2(b)).
de Miranda et al. (2010) measured BC concentrations using the optical reflectance method in six large Brazilian capitals, and found mean concentrations in the ranges 2.30-7.10µg m -3 in summer and 4.00-13.10µg m -3 in winter, with the largest concentrations in São Paulo city.Their results are difficult to compare with this study since they used another measurement technique and there is no detailed information provided by the site descriptions (e.g., if the sites were street canyon or urban background).Backman et al. (2012) reported median hourly BC concentrations between 2.00 and 5.40 µg m -3 at an urban site in São Paulo city in the period October 2010-January 2011.A study by Krecl et al. (2011) in Stockholm -with the same filter-based optical technique used here-reported mean concentrations of 5.39 µg m -3 in a street canyon site, 1.13 µg m -3 in a rooftop station and 0.36 µg m -3 in a rural area with minor anthropogenic influence.The street canyon and rooftop concentrations compare well with our results.

Effects of Meteorology on BC Concentrations
We used cluster analysis to group days with similar meteorological conditions and to relate BC diurnal patterns to meteorological variables.We applied Ward's minimum variance hierarchical clustering method (Ward, 1963) on the squared Euclidean distances between hourly data of T air , RH and WS.Each dataset was clustered into three groups of days with similar meteorological patterns in terms of T air and RH, which we refer to as period 1 (P1), period 2 (P2) and so forth up to period 9 (P9) (Fig. 3).The diurnal patterns of WS showed little differences within each period and had a weak relationship with BC diurnal cycle.We show the P95 concentrations because the differences among the periods stand out for this statistical indicator, as shown in Fig. 2(b).
During UTF dry , the mean hourly RH ranged from 23 to 90% and the mean hourly T air from 15 to 30°C (Fig. 3(a)).P3 was characterized by milder temperatures (max T air = 23°C) and higher RH (between 65 and 90%), while P2 and P3 were hotter and drier.The combination of dry (minimum RH = 23%) and warmer conditions (maximum T air = 30°C) triggered elevated BC concentrations especially in period P2 not only in the evening (95 th percentile = 18.28 µg m -3 at 20:00) but also in the morning.The first morning peak  (between 1:00 and 3:00) can be attributed to the welldocumented effect of shallower nocturnal boundary layer that leads to the build-up of particles near the surface, while the second morning is related to anthropogenic activities at the campus.A decline in wind speed was also observed from 3.8 m s -1 at 15:00 to 2.4 m s -1 at 20:00 (not shown), coincident with BC enhancement.A similar night peak was observed in UTF wet (P6), however, the smaller mean hourly RH range (51 to 97%) and more frequent rainfall may have not favored the occurrence of local fires, in spite of the high T air attained in this period (up to 32°C).No clear pattern was observed in the rooftop measurements.In fact, relatively larger BC concentrations seemed to be related to wet (approx.90%) and cool (approx.21°C) conditions observed in P9.Despite the low RH and high T air observed in P8, the largest BC concentrations in the canyon occur in rush hours in intermediate T air and RH conditions (P7).
Modelling (Textor et al., 2006) and observational (Taylor et al., 2014) studies showed that wet deposition (both nucleation and impaction) dominates the removal of BC from the atmosphere, and for typical BC size distributions impaction scavenging favors smaller BC (Aitken mode), while nucleation scavenging favors larger BC (accumulation mode) (Jacobson, 2003).Fig. 4 shows the time series of BC concentrations for all datasets along with the hourly rainfall.
The BC concentrations during UTF dry were consistently higher than UTF wet , with a dome-like pattern in periods of sustained large concentrations that build up over a few days (e.g., between 16 and 25 August 2014, Fig. 4(a)), superimposed by sporadic peaks.These periods were disturbed by rainfall events, when BC particles were either rained out abruptly (e.g., a total of 23.6 mm on 13 August) or kept from accumulating in the atmosphere (e.g., a total of 35.3 mm on 19 and 20 September).December 2014 and January 2015 were relatively wetter and BC levels were kept low due to the frequent rainfall, except between 15 and 21 December when the dry spell (total of 1.8 mm in the period) contributed to enhanced BC concentrations (Fig. 4(b)).Due to equipment malfunction in the canyon, we did not obtain concurrent BC data during the most significant rainfall events, which explains why a direct relationship between rainfall and BC removal was not established (Fig. 4(c)).The rooftop BC concentrations decreased during or right after the few rainfall events captured towards the end of November 2014 (Fig. 4(d)).

BC Concentrations in Different Seasons: Contributions of Local and LRT Smoke
In southern Brazil, air quality is linked to alternating clean southerly flows and polluted northerly flows that

A
Targino and Krecl, Aerosol and Air Quality Research, 16: 125-137, 2016 133 bring biomass burning smoke from savannahs and forest fires of the central and northern regions.BC levels measured at UTF respond to local backyard burning, LRT of smoke and occurrence of rainfall, which contribute to alternating clean and polluted periods (Figs. 4(a) and 4(b)).
The polluted season occurs in winter and spring and is related to the lingering blocking highs which prevent the rain-bearing southerly cold fronts from penetrating the region (Satyamurti et al., 1998).Prins et al. (1998) found that the majority of the biomass fires observed in South America between 0 to 40°S and 35 to 75°W occurred in August and September, and recent data reported 444,059 fire foci nationwide in August and September 2014 (INPE, 2015).The onset of the rainy season in October diminishes the fire foci (58,290 in the period December 2014-January 2015) and the increase of the aerosol optical depth is more related to smoke bursts due to local emissions (Rosário et al., 2013).We chose two subperiods within UTF dry and UTF wet , and combined location of fire foci in South America (INPE, 2015) with 120-h air mass backtrajectories calculated with HYSPLIT (Hybrid Single-Particle Lagrangian Integrated Trajectory) model (Draxler and Rolph, 2011) arriving in Londrina at 500 m above terrain level with 1-hour interval.Other trajectories approach from south and west crossing burning areas in Argentina and Bolivia.
The number of fire foci in Fig. 5(b) is 17,749 and most of them occur in northern Brazil, with little influence over Londrina.The trajectories in this subperiod are longer and faster, with two distinct groups -from the south Atlantic and northeastern Brazil -and travelling over relatively smaller burning areas.In summary, the mean (± SD) and median BC concentrations for the dry period subperiod are 1.70 ± 1.13 and 1.47 µg m -3 , and for the wet subperiod 0.46 ± 0.55 and 0.35 µg m -3 .To help identify local sources and which BC concentration percentiles are more affected, we calculated the pollution rose by using 10-min WD measurements and concurrent BC concentration data for λ = 370 and 880 nm (except for the canyon site where we show results for λ = 880 nm only) averaged over 22.5-degree intervals.
Aerosols emitted by the combustion of biomass and fossil fuel absorb radiation in the spectral range from the ultraviolet to the infrared.However, organic compounds in biomass have stronger absorption spectral dependence and largely contribute to absorption at shorter wavelength (e.g., 370 nm), while vehicle exhaust particles exhibit relatively weak wavelength dependence, that is, the absorption across the spectrum varies little.For example, Kirchstetter et al. (2004) found that the absorption varied from λ -1 for motor vehicle aerosols to λ -2.5 for biomass burning aerosols.Figs.6(a) and 6(b) illustrate the BC concentration rose for the median and P95 within UTF dry and UTF wet , respectively.The median BC rose for both wavelengths within UTF wet shows that the emission sources have little direction dependence and concentrations do not overpass 0.5 µg m -3 .During UTF dry , sources in the wind segment 270-315°, 180-215° and 0-45° produced median concentrations larger at 370 nm (between 1.64 and 3.80 µg m -3 ) than at 880 nm (between 1.21 and 2.70 µg m -3 ), which indicates a wavelength dependence of light absorption by the aerosol, typical of smoke from biomass sources.This suggests that at the campus smoke from biomass dominates over smoke from fossil fuel.We clustered the backtrajectories within UTF dry according to the length, pathway and height (not shown).The clustering yielded 5 groups, with 68% of the airmasses approaching from north-northeast and 33% from south which, as shown in Fig. 5, are sectors with high occurrence of fire.Hence, this median pattern may represent the regional LRT contribution and the sectors are related to the most common pathways of air masses in the period.Contributions from the 270-315° sector affect the P95 concentration rose within both UTF dry and UTF wet with larger concentrations at λ = 370 nm (max.19.76 µg m -3 within UTF dry ) than at λ = 880 nm (max.14.09 µg m -3 within UTF dry ).These large concentrations indicate that contributions from local biomass smoke have a preferential direction, are intermitted and affect the highest percentiles in both seasons.Visual inspection showed that when local fires occur, the smoke raises northwest of the campus.The WS and WD roses (not shown) revealed that winds blowing from this sector are neither strong nor frequent, however they contribute to significant sporadic enhancement of BC concentrations at the university campus.At the urban background site, P50 has a weak BC enhancement at both λ = 370 and 880 nm for the same wind sectors observed for P50 at the university campus, which suggests that regional LRT also affects this site (Fig. 6(c)).It is interesting to note that while P50 concentrations increased mostly at λ = 370 nm, indicating a larger contribution from biomass smoke, P95 concentrations at λ = 370 and 880 nm do not diverge (Fig. 6(d)).This may be interpreted as a contribution mainly from motor traffic, since the weak dependence of the absorption coefficient on this type of aerosol yields comparable BC concentrations at both wavelengths.The BC concentrations at the canyon site are larger in the W, SE and S sectors.As previously mentioned, WD perpendicular to the street axis triggers recirculation vortices and creates localized air pollutant patterns in canyons.In this study, winds in the W, SW, SE and S directions account for 51% of the dataset and contributed to larger P50 and P95 BC concentrations in the canyon.We attribute the BC enhancements to: i) transport of aerosols from a heavily-trafficked road (Minas Gerais street) that runs perpendicular to the canyon and west of the site, and ii) recirculation vortices that create higher BC concentrations on the leeward side (south side).

CONCLUSIONS
We investigated the BC concentrations at three sites of a mid-sized city in southern Brazil and observed a large spatiotemporal variability in the dataset.The street canyon site had the largest mean and standard deviation (3.00 ± 2.35 µg m -3 ) and 95 th percentile concentration of 7.58 µg m -3 , with peaks at 7:00 and 18:00, coinciding with the traffic rush hours.The concentrations at the canyon site were larger for wind in the W, SE and S sectors, which represents the effect of recirculation vortices inside the canyon and the transport from traffic aerosols in a busy street located west of the site.The BC concentrations at the suburban site were highly variable and dependent on both remote and local sources.A detailed investigation into two subperiods within the dry and wet periods showed that the mean BC concentrations at the suburban site were governed not only by backyard biomass burning but also by long-range transport of biomass smoke from other regions of Brazil and neighboring countries, from August to September.The BC diurnal cycle at this site were composed of a mean cycle that can be attributed to the frequent transport of smoke from wildfire areas, and of a 95 th percentile cycle that peaks in the evening in response to sporadic events of backyard burning in dry and hot weather conditions.Larger BC concentrations when recorded at the urban rooftop were attributed to emissions from motor traffic in busy streets adjacent to the site.This study reinforces the need of finer BC monitoring network to capture patterns of BC concentrations in a city affected by motor traffic, local and LRT plumes, and raises awareness on the deterioration of air quality and health damage caused by local waste burning.Our results also suggest that, while air pollution due to LRT biomass smoke is more difficult to abate, targeting local backyard burning and traffic would lead to a depletion of BC concentrations in the city.

Fig. 1 .
Fig. 1.Map of Londrina and location of the monitoring sites.

Fig. 2 .
Fig. 2. (a) Mean diurnal cycle of BC concentrations (880 nm wavelength) and (b) diurnal cycles of mean and 95 th percentile of BC for weekdays and weekends for the four datasets.

Fig. 3 .
Fig. 3. Left column: clusters of diurnal cycle of air temperature (solid lines) and RH humidity (dotted lines).Right column: diurnal cycle of 95 th percentile of BC concentrations observed in each cluster.University campus, dry period (ab); University campus, wet period (c-d); rooftop (e-f); street canyon (g-h).

Fig. 5
shows the fire foci (red dots) in Argentina, Bolivia, Brazil, Paraguay and Uruguay, and the backtrajectories (blue lines) in the dry subperiod (16 to 25 August 2014, Fig. 5(a)) and wet subperiod (03 to 09 December 2014, Fig. 5(b)).Fig. 5(a) shows 125,453 fire foci and most trajectories following a northeastern path with anticyclonic curvature and travelling over areas with large concentrations of foci.

Fig. 6 .
Fig. 6.Median (left column) and P95 (right column) BC concentration roses (µg m -3 ) for UTF (a) and (b); rooftop (c) and (d); and street canyon (e) and (f).Note that the scales vary in the plots to facilitate visualization of small concentrations.

Table 1 .
Summary of the operating conditions of the aethalometers.i,last is the time of the last measurement data for filter spot i, and t i+1,first is the time of the first data for the next filter spot

Table 2 .
Mean k values and standard deviation (= 880 nm) calculated for each instrument and sampling period.

Table 3 .
Summary of the meteorological conditions.
a Mean values.b Relative frequency of predominant wind direction sector.c Cumulative precipitation in the period and daily maximum.d Number of days with precipitation larger than 1 mm.

Table 4 .
Descriptive statistics for 10-min averaged BC data (λ = 880 nm).SD stands for standard deviation and n is the number of valid samples.