Prevalence of Freshly Generated Particles during Pollution Episodes in Santiago de Chile

A winter campaign was carried out in Santiago de Chile the year 2012 in two urban sites that can be considered representative for most of the city in order to characterize formation of primary and secondary PM1.0 during episodes. One site is located in the campus of the University of Santiago and measurements were carried out with an Aerosol Chemical Speciation Monitor and a black carbon monitor. Another site is located in a large park, about 2 km south-east of the first site, measurements of CO, NOx, SO2, O3 were done in this site. A noticeable increase in most of the primary components of PM1.0 (black carbon and organics) and primary gases (CO and NO) was observed during days in which the average PM1.0 concentration was higher than 50 μg m (episode). A small increase or no change was observed in the secondary pollutants (NH4, NO3, SO4 and NO2) at night during these episodes. Positive Matrix Factorization was used to extract four components from the ACSM data: hydrocarbon-like organic aerosol (HOA), biomass burning OA (BBOA), low volatility oxygenated OA (LV-OOA) and semi-volatile oxygenated OA (SV-OOA). The freshly generated components (HOA and BBOA) showed a clear increase at night during episodes, while the aged fraction of organic aerosol (LV-OOA and SVOOA) showed a smaller increase or a decrease at night during episodes. Correlation of HOA and BBOA components with primary pollutants was also high, indicating that freshly created aerosols (HOA, BBOA and BC) are in large part responsible for the increase in pollution at night during episodes in Santiago de Chile.


INTRODUCTION
Santiago, Chile's capital is a large city (6 million of people) in South America that is surrounded by hills and experiences adverse meteorological conditions that favor high pollution during winter.Currently, PM 2.5 annual concentration is over 2.2 times the Chilean standard (20 µg m -3 ), but the highest PM 2.5 levels are seen during winter (Koutrakis et al., 2005).During this period, it is also common to observe an increase in the number of children's respiratory decease's as well as other health problems, following a pollution event (Ilabaca et al., 1999).As a consequence, the government developed a number of control strategies to help reduce pollution.In 1995, the "Environmental-Base Law" was passed and directed the National Commission for the Environment (Conama) to make up a pollution-control plan for Santiago and its surroundings.This plan, which was completed on July 25, 1997, established specific emissions reduction targets for the most common pollutants such as particulate matter, PM 10 , ozone, O 3 , nitrogen oxides, NO x , sulfur dioxide, SO 2 and CO.This plan has been quite successful because PM 10 has decreased from 90 µg m -3 in 1997 to 60 µg m -3 in 2001 (Koutrakis et al., 2005).Since that year, the PM 10 fraction has remained approximately constant.However, this plan did not address the fine fraction of particulate matter, which during winter reaches levels much higher than the WHO recommendation (Koutrakis et al., 2005).In May of 2011 new regulations were enacted to control the fine fraction of particulate matter, PM 2.5 .This fraction is much harder to reduce because a large part of it is of secondary origin (Carbone et al., 2013) and its concentration is not directly related to the sources, but to the chemical reactions that occur in the atmosphere.A variable part of PM 2.5 corresponds to black carbon (BC) and brown carbon (BrC) (Kirchstetter and Novakov, 2004), which strongly absorb light across the spectrum and are believed to be the second most important pollutant associated to global warming (Ramanathan and Carmichael, 2008;Bond et al., 2013;Liu et al., 2015).Both types of carbon are generated in large quantities in cities, and it has been recently shown (Liu et al., 2015) that their absorption properties depend on the mixing state, morphology and age as well as being source and regionally specific.
Fine particulate matter has several classifications that depend on its origin, the instrument that is used to measure it, or its reactions in the atmosphere.Non-refractory PM 2.5 is the fraction that volatilizes at temperatures ~lower than 600°C, and corresponds to organics, nitrates, sulfates, ammonium, etc. Refractory PM 2.5 is composed mostly of black carbon.PM 2.5 can also be classified as organic or inorganic and each fraction can be of primary or secondary origin.NH 4 , NO 3 , Chl and SO 4 correspond to the inorganic fraction, while the organic fraction (OA) can be divided into primary and secondary OA (POA and SOA, respectively).Carbone et al. (2013) estimated that the contribution of OA to fine particle mass, can reach up to 59% in spring in Santiago using an Aerosol Chemical Speciation Monitor (ACSM).Their data were analyzed by applying positive matrix factorization (PMF) to the organic mass spectra.They observed a change in the composition of OA during the transition from winter to spring.The emissions from primary sources, such as vehicle and biomass burning, decreased in the period leading to spring, whereas the amount of oxygenated organic aerosol increased over the same time.The organic fraction (OA) of PM 2.5 in Santiago has been studied for short periods of time since 2003 (Conama, 2003(Conama, , 2005;;MMA, 2011), but the results of the studies have not been correlated to the sources or distribution of OA in the city.It is also known that in order to implement efficient policies for improvement of the air quality on the city it is important to quantify the amount of fine particle matter that is freshly emitted (primary) and the amount that comes from chemical reactions in the atmosphere (secondary).
A frequent problem in Santiago de Chile is the occurrence of pollution episodes with PM 2.5 peaks of 200 µg m -3 or more (Koutrakis et al., 2005;Toro et al., 2014).These episodes occur at night and are related to adverse meteorological conditions such as thermal inversion and low wind speed (Rutland and Garreaud, 1995;Gramsch et al., 2000) as well as the location of the episode (Gramsch et al., 2006).It is not clear whether the episodes are related to a build-up of secondary PM 2.5 components or to freshly emitted pollutants (Gramsch et al., 2014).It is also well known that Santiago has a diurnal-nocturnal, valley-mountain wind cycle and as a result, the air mass can remain an average 2-3 days trapped on the city with the consequent enhancement of the secondary PM 2.5 fraction.
In this work, all major components of fine particulate matter, gases and meteorological parameters have been measured for a period of four months during fall and winter in Santiago in order to get a better understanding of the temporal variation and sources of PM 2.5 during episodes.Although, the normative always refers to PM 2.5 size fraction, it is well known that most anthropogenic PM 2.5 is contained in PM 1.0 (Seinfeld and Pandis, 2006).Black carbon particles are also known to have sizes below ~0.3 µm, so in the remaining of the manuscript PM 1.0 is going to be used, because this is the quantity measured by the ACSM and carbon monitor.

EXPERIMENTAL SECTION
Measurements were performed in two sites in downtown Santiago (Usach and Parque O'Higgins) in order to include in the study data from instruments that are not available at both sites.However, previous studies (Gramsch et al., 2006) have determined that Parque O'Higgins and Usach are located in an area that has similar topographical, meteorological and pollution characteristics.The map in Fig. 1 shows the location of the sampling sites and the predominant wind direction is indicated in the figure with an arrow.

Usach Sampling Site
The site is located in the middle of the campus of the University of Santiago, about 500 m north of the main street (Alameda) and 200 m west of Matucana Street.Alameda is the largest street in Santiago and has a flux of about 60,000 vehicles per day (Gramsch et al., 2013), Matucana is another large street with a flux of about 30,000 vehicles per day.The university campus is located approximately in the middle of the city, about 3 km west of downtown.The area has a lot of commercial and retail activity, but it is also a residential area.Because the site does not have direct influence from pollutions sources and is located in the middle of the city, it can be considered as an urban background station.
In this site, between 9 March and 5 July 2012 continuous measurements of non-refractory PM 1.0 (sulfate, SO 4 , nitrate, NO 3 , chloride, Chl, ammonium, NH4, and organics) were done with an Aerosol Chemical Speciation Monitor (ACSM, Aerodyne Research, Inc.), black carbon was measured with an optical monitor developed at the University (Gramsch et al., 2000).This monitor used a filter to capture particulate matter and the absorption of the light is used to determine the amount of black carbon.The sample inlet of the ACSM was located about 3 m above ground and the sample inlet of the carbon monitor was located about 10 m above ground, in the roof of a building.

Parque O'Higgins Sampling Site
It is located in a large park about 2 km south of the city center and 1 km west of a major highway with traffic of about 60,000 vehicles per day.The station does not have direct influence from pollutions sources, thus it can be considered as an urban background station.The area has a mixture of houses, retail and light industries (machine shops, auto repair shops, furniture manufacturing shops, etc.).This site belongs to the Macam Network (Gramsch et al., 2006) and it is operated by the Ministry of the Environment.This station has monitors for PM 10 , PM 2.5 , O 3 , CO, SO 2 and meteorological parameters.

ACSM Instrument
The Aerosol Chemical Speciation Monitor (ACSM) is a relatively new instrument that has the important ability to sample several non-refractory aerosol components, such as nitrate, sulfate, ammonium, chloride and organics (Ng et al, 2011) in real time.These components change during the day and can give important information with respect to the sources of fine particulate material.The ACSM has an aerodynamic lens to sample and focus PM 1.0 particles with a 50% efficiency.The particles are transmitted through vacuum chambers into a hot oven that vaporizes the particles, which is usually kept at 600°C.The vapor that comes out of the oven is ionized with a 70 eV electron impactor and is directed into a quadrupole mass spectrometer.The spectrometer is a Prisma Plus System from Pfeiffer Vacuum that has an overall sensitivity of 6 × 10 -4 amps/mbar.The detection chamber has an internal calibration standard consisting on an effusive source of naphthalene that enters the chamber via a 1 µm pin hole.The parent peak for naphthalene is m/z = 128 and provides a continuous internal standard for calibrating the mass to charge ratios of the measured ions.Therefore naphthalene is always present in the mass spectra.
In addition to the continuous calibration, the response factor (RF) of the ACSM is obtained at the beginning of the campaign using ammonium nitrate aerosol using the procedure described by Ng et al. (2011).An ultrasonic nebulizer was used for primary aerosol generation, with silica gel diffusion drier, a Hauke-type differential mobility analyzer and a condensation particle counter, model 3010 from TSI, Inc.The nebulizer was used to generate 300 nm ammonium nitrate aerosol particles.The mass concentration of the injected particles was calculated using data from the counter, the known diameter and density of the particles, assuming that they are spherical.Comparing the calculated ammonium nitrate mass concentration and the values provided by the instrument, the calibration can be performed.Varying the dilution of the generated aerosol the number concentration could be varied between 10 and 1000 1 cm -3 , which corresponds to a mass concentration between 0.15-15 µg m -3 of nitrate.

Black Carbon Monitor
The monitor used in this study (SIMCA) employs a variation of the integrating plate method (Lin et al., 1973;Horvath, 1993;Gramsch et al., 2000) to measure the absorption coefficient of light in the air.This coefficient is multiplied by the mass absorption coefficient (Horvath, 1993, and references therein) to obtain carbon concentration.The SIMCA is made up of a head containing a filter, two light-emitting diodes (LEDs), two photodetectors, and amplifier electronics.A computer controls a pump, the LEDs and photodetectors through an interface box.

Gases and Meteorological Data Measurement
Continuous measurements of PM 2.5 were done with a beta attenuation monitor (Thermo Scientific 5014i).The gases measured are CO (Monitor Labs 9830), NO, NO 2 , O 3 (API T204), SO 2 (API T100).Meteorological parameters were measured with a NovaLinx 110-WS-25 weather station.

Positive Matrix Factorization Analysis
In order to further investigate the components of the organic fraction, Positive Matrix Factorization (PMF), developed by Paatero and Tapper, 1994 was used in the robust mode PMF evaluation panel version 2.4 developed by Ulbrich et al. (2009).In this technique, the bilinear equation: is solved, where x ij are the concentrations of j species in i samples, p are factors with constant source profiles.PMF minimizes the summed least squares errors of the fit weighted with the errors of each data point, with the constraint of having solutions with non-negative values.
The input to the equation is the measured concentration of organic aerosols with their respective uncertainties, where the rows represent time series and the columns represent the mass-to-charge ratios (m/z's) (Ulbrich et al, 2009).Before applying the PMF analysis a pretreatment of the input data was performed recommended as described by Ulbrich et al. (2009) PMF analysis gives sporadic spikes of individual variables (m/z's).In order to minimize this error, the spikes of individual m/z's were downweighed 100 times, thus decreasing their importance inside the fit.

Wind Pattern
The behavior of pollution in Santiago is very dependent on the wind pattern, which in turn is dependent on the topography and location within the city (Gramsch et al., 2006).In the afternoon, in the eastern part of the city the wind blows from west to east and at night, the direction reverses to an east to west direction.In the center and western part of Santiago, there are higher wind speeds in the afternoon and because the wind comes from the west, it brings clean air into this part of the city and the concentrations of PM 2.5 and PM 10 reach their lower levels.However, at night and early morning the wind that comes from the mountain (east) does not reach the center or western part of Santiago and the atmospheric stability is very high.The wind speed during the afternoon reaches ~3 m s -1 in summer and only ~1 m s -1 in winter.At night the wind speed is even lower, reaching at most 0.5 m s -1 .The predominant wind direction during the afternoon is indicated with an arrow in Fig. 1.It can be seen that at this time of day both sampling sites receive clean air from the west and there is little transport of pollution from one site to the other.

Time Series
Measurements were performed during a period that was long enough to capture several high pollution episodes during fall and winter.A comparison between periods with high and low concentrations allows understanding how the high pollution episodes are developed.The time series for the combined ACSM measurements (OC + NO 3 + SO 4 + NH 4 + Chl) and BC at Usach site is shown in Fig. 2. The same figure shows PM 2.5 measured with a Beta monitor in Parque O'Higgins site.Although the monitors are not colocated, and the size fractions are different, both curves show a similar trend, with a correlation coefficient R 2 = 0.50.On average the ACSM + BC corresponds to 92% of the PM 2.5 measured by the Beta monitor.As mentioned before, the ACSM sample particles below 1.0 µm, and black carbon is mostly composed of particles with diameters below ~0.3 µm (Seinfeld and Pandis, 2006), so these measurements The largest fraction measured by the ACSM at Usach site corresponds to organics with an average concentration of 15.4 µg m -3 followed by BC with 7.0 µg m -3 for the period (Fig. 3(a)).These two fractions have a relatively large correlation of R 2 = 0.51, n = 2392, because some of the sources are the same.BC is of primary origin and is emitted by vehicles, diesel engines, wood burning, cooking, etc. Organic aerosols are also emitted mostly by the same sources (although with different emission factors) but it has also secondary origin.In the figure, it can also be seen that the concentration of BC and organics increases as winter approaches.This fact can be explained because in winter, the wind speed is lower and the temperature inversion is stronger.The time series for NH 4 and NO 3 is shown in Fig. 3(b).The average concentration for NO 3 is 5.2 µg m -3 and 2.7 µg m -3 for NH 4 between March and July.The correlation coefficient between these components is very high R 2 = 0.85.Nitrate forms several compounds with other elements in the atmosphere (KNO 3 , HNO 3 , NaNO 3 , NH 4 NO 3 , Ca(NO 3 ) 2 , etc.), however, the high correlation between NH 4 and NO 3 indicates that most of the nitrate in the air of Santiago is in the form of ammonium nitrate (NH 4 NO 3 ).A small decrease in the trend is seen for NH 4 and NO 3 as winter approaches.The most likely reason for this decrease, is that NO 3 and NH 4 are of secondary origin and need solar radiation to be formed in the atmosphere, thus as winter approaches, less radiation is available to form these compounds.
Sulfate and chloride are shown in Fig. 3(c).Both have very low concentrations compared to the other compounds, with an average of 0.98 µg m -3 for SO 4 and 0.61 µg m -3 for Chl.SO 4 is only 3% of the total amount, which is much lower than other cities.For instance, in Beijing, SO 4 amounts to 18% of the total (Sun et al, 2012), 28.8% in Queens, NY (Ng et al., 2010), 19% in Atlanta (Budisulistiorini et al., 2013).
SO 4 is formed in the atmosphere from SO 2 , which in turn can be emitted by diesel vehicles, electric generators, power plants, cooper smelters, etc.In a city the most important source are diesel vehicles, however, in Santiago, the diesel fuel used has only 15 ppm of sulfur content and there are very few large industries that can emit SO 2 .In contrast, outside the city, there are several copper smelters (Caletones smelter is located 72 km south of Santiago, Chagres smelter is located 80 km north and Ventanas smelter is located 116 km north-west) that generate large quantities of SO 2 which can reach the city if the winds are favorable (see discussion in Olivares et al., 2002, Carbone et al., 2013).The correlation between SO 2 and SO 4 in the same period is R 2 = 0.04, which indicates that these contaminants do not have the same sources and it is also an indication that SO 4 may come from outside of the city.Fig. 3(d) shows that there are several large SO 4 peaks that are unrelated to SO 2 or any of the other contaminants which also indicate that SO 4 may come from outside of the city.
The composition of PM 1.0 and PM 2.5 obtained in different studies in Santiago is shown in Table 1.The only components that seem to have increased in proportion after 2005 are organic and black carbon.This increase is consistent with the growth in the number of vehicles in Santiago de Chile (Moreno et al., 2010).It has to be noted that measurements of organic carbon in 2003 and 2005 were performed with thermo-optical methods (Ambient Carbon Particulate Monitor 5400, from Rupprecht and Patashnick Co.).The largest fraction of PM 1.0 is organics, followed by black carbon, nitrate and ammonia; this is shown in Fig. 4 for the March-July period.Organics corresponds to 48% of the total followed by BC with 22%.In Santiago, like in most cities, organics is the largest fraction of PM 2.5 (Ng et al., 2010;Sun et al., 2012;Budisulistiorini et al., 2013).

Pollution Behavior during Episodes
In order to understand the behavior of the atmosphere during days with high PM 2.5 concentration, the measurement period has been divided in days with and without episodes.An episode is defined as a day in which the average PM 2.5 (measured at Parque O'Higgins) is higher than 50 µg m -3 .With this definition, 8 episodes and 105 non-episodes were found.The diurnal profile for the whole period (March 10-July 4, 2012) separated for days with and without episodes is shown in Fig. 5 for several pollutants.In the figure, it is clear that for all pollutants the concentration is higher in days with episodes and during these days, the wind speed is lower and the temperature difference during the day is higher.During non-episode days, the wind direction in the afternoon is south-west, as indicated with an arrow in Fig. 1, but during episodes, the wind speed is so low, that there is no predominant direction.These meteorological conditions favor building up of contaminants (Schaefer et al., 2006;Gramsch et al., 2014) therefore increasing pollution during episodes.Another feature that can be seen in the plots is that the increase in PM 2.5 is larger at night (8 pm-5 am) than during the day.It is well known that pollution episodes in Santiago (Gramsch et al., 2000) and many other cities (Watson et al., 2002, Atkinson et al., 1986, Wallace et al., 2009) occur at night during winter, because the inversion layer height is lower, wind speed is lower and, in many places wood burning used intensively for space heating.
Fig. 5 also shows that the concentration of some contaminants decreases or has a small increase at night compared to the day.If the curves with and without episodes  are compared, it can be seen that during episodes, organics, BC, CO, NO and SO 2 have an increase at night that is much larger than the increase during the day.On the other hand, NH 4 , SO 4 , NO 3 and NO 2 have an increase at night that is similar or only slightly larger than the increase during the rest of the day.
To quantify the difference between day and night, the first column (Day) in Table 2 shows the average of all measurements between 6 am-7 pm. with episodes divided by the same average without episodes.The second column (Night) shows the average between 8 pm-5 am. with episodes divided by the same average without episodes.It can be seen that this ratio is much larger for NO, CO and BC, i.e., that have the highest increase at night during episodes.In most cities, these pollutants are emitted by vehicles but also by wood stoves (AP-42, 2009).CO is released in any incomplete combustion; in particular wood stoves have very inhomogeneous combustion that favors CO emissions.Organic pollutants are emitted by vehicles and wood stoves but also by several other sources.SO 2 in Santiago is emitted by diesel generators or industrial burners, but not from vehicles.Thus, it can be seen that the species that have a high increase at night during episodes are the primary pollutants and the species that have a smaller increase or decrease these are secondary pollutants.
A six factor statistical analysis using positive matrix factorization (PMF, Paatero and Tapper, 1997) has been performed with the ACSM data to separate the main components of organic aerosol.Some of these factors were recombined to obtain a 4 factor solution.The sum of the squares of the residuals (Q-value) was minimized and used to choose the number of sources that can be extracted from the data.A comparison of the diurnal cycles of the factors was also used to determine which factors could be recombined.Finally, a comparison of the mass spectra with previously published data in Santiago (Carbone et al., 2013) was also performed.The mass spectra of the six factors are shown in Fig. 6.The first factor is characterized by having several m/z peaks of 16, 27, 29, 41, 43, 55, 57, 66, 69, 71, 81, 83, 91, 95, 97.This spectrum is typical of molecules arising from vehicular exhaust, like akanes, alkenes, etc. (Zhang, et al., 2005).The second spectrum in Fig. 6 resembles the first because several of the peaks are the same.In addition its diurnal profile is somehow similar, as shown in Fig. 7(a).Factor 2 shows an increase during the morning rush hour, both curves show a decrease during the afternoon and a second increase during the evening.The correlation coefficient between factor 1 and 2 was R 2 = 0.67.The daily profiles are also very similar to CO, NO and black carbon (see Fig. 5), typical of vehicular emissions.The mass spectrum is also similar to what was found by Ng et al. (2010) for Hydrocarbon-Like Organic Aerosol (HOA).Thus, the first two factors were combined into one and associated with HOA.Previous studies have also found that HOA correlates well with several primary gases such as CO and NO (Zhang et al., 2005;Lanz et al., 2007) and can be considered a tracer for primary pollutants from combustion.In this study, the correlation between HOA and CO is R 2 = 0.45, n = 2688 and between HOA and NO is R 2 = 0.37, n = 2688.These values are not as high as what was found in the previous references and the reason for the lower correlation is most likely that HOA was measured at the Usach site and CO and NO were measured in Parque O'Higgins site.
The third factor in Fig. 6 has m/z's peaks in 27, 29, 41, 43, 55.This mass spectrum has been associated before to cooking (Lanz et al., 2007, Allan et al., 2010) which in Santiago could arise from household cooking, restaurants and cooking from street vendors.Peaks 41, 43 and 55 arise from oxygen-containing ions such as from fatty acids (Mohr et al., 2009).The daily profile of this factor is also similar to factor 4, as shown in Fig. 7(b).Factor 4 has peaks in 15, 29,31,41,42,55,57,60,73.Previous studies have associated the m/z's peaks 29, 60 and 73 to wood burning because it correlates well with levoglucosan, acetonitrile and potassium (Lanz et al., 2007;Aiken et al., 2009).However, factor 4 also has m/z's peaks at 29, 41 and 55 that can also be associated to cooking (Lanz et al., 2007;Allan et al., 2010), thus this factor has contributions from two (or more) types of biomass burning.Because the daily profile of these factors is similar, and the Pearson correlation coefficient between factor 3 and 4 was R 2 = 0.33, they have been combined into one and it has been labeled Biomass Burning Organic Aerosol (BBOA).This factor correlates well with primary pollutants, for instance, correlation between BBOA and CO is R 2 = 0.50, and with NO is R 2 = 0.46.In a city, most CO and NO are emitted by vehicles, but wood burning also generates these compounds (AP-42, 2009).The correlation between BBOA and organics is R 2 = 0.67 and between BBOA and BC is R 2 = 0.42.These last correlations are consistent with the fact that wood burning generates more organics than BC.
The fifth factor has mass spectrum peaks in 18, 29, 41 and 44 and has been labeled semi-volatile oxygenated organic aerosol (SV-OOA).Previous studies have found high correlation between this factor and secondary aerosol such as nitrate or sulfate (Lanz et al., 2007;Ulbrich et al., 2009).In this work, it has been found that this factor has high correlation with nitrate (R 2 = 0.4) and ammonium (R 2 = 0.53), but very low correlation with sulfate (R 2 = 0.09).
The last factor has a mass spectrum dominated by m/z 44.This is the most oxygenated fraction of the organic aerosol (OA) and it is related to the CO 2 + fragment from thermal decarboxylation of organic acid groups (Alfarra, 2004).This more oxidized factor is referred as LV-OOA (low volatility OOA; Ng et al., 2010) and has been found to correlate well with secondary pollutants like NO 2 , NH 4 , SO 4 , etc. (de Gouw et al., 2005;Zhang et al., 2005;Volkamer et al., 2006).
Particulate sulfate is usually found to have high correlation with OOA.Lanz et al. (2007) found that the correlation between sulfate and OOA was R 2 = 0.51 in Zurich, but in Santiago, the correlation between sulfate and LV-OOA is R 2 = 0.09, n = 2688, and between sulfate and SV-OOA is R 2 = 0.01, n = 2688.In either case there is no correlation.As mentioned before, the most probably reason for this fact is that sulfate is generated by primary pollutants that come from outside the city, while most OOA is generated by primary pollutants emitted in the city.NO 2 is also generally well correlated to OOA, but in this study, the correlation with LV-OOA is R 2 = 0.06 and the correlation with SV-OOA is R 2 = 0.22, that is, NO 2 is slightly better correlated with newer OOA.
In order to visualize the sources of pollution, the four OA components been plotted as function of time in Fig. 8.It can be seen that the lowest total concentration is in March, which is the last month of summer and increases as winter approaches.The main reason for the higher concentration in winter is the lower temperature which forms stronger inversions and the lower wind speed which  prevents dispersion of contaminants.Table 3 shows the average of the meteorological parameters for each month of the measurement period; a clear decrease in wind speed and temperature is seen.Fig. 8 also shows that all factors grow from March to June, but BBOA is the factor that increases dramatically, from 0.34 µg m -3 (3.3%) in March to 4.64 µg m -3 (45.8%) in June.The growth in LV-OOA, SV-OOA and HOA is most likely related to a smaller atmospheric boundary layer, but BBOA must be related to an increase in the use of wood stoves.In Santiago, wood burning is used mainly for space heating, which increases as temperature decreases.
The daily profile of the four factors has been calculated for the whole period.As before, data has been separated in days with and without episodes.Results are shown in Fig. 9.The plot shows that the primary organic components, HOA and BBOA, have higher concentrations values at night during episodes.If the day average is calculated from 6 am to 7 pm and the night average is calculated from 8 pm to 5 am of the next day, then the night average is 88% higher  for HOA and 98% higher for BBOA during episodes.On the other hand, the older organic components, SV-OOA and LV-OOA show a smaller increase or even decrease at night during episodes.The increase in concentrations for BBOA and HOA is significant at 0.01 confidence level, according the t-test.LV-OOA, which is the oldest organic component, even shows a decrease at night during episodes.
For SV-OOA, the t-test shows that the means of the night and day averages are equal at 0.01 confidence level.For LV-OOA the t-test shows a decrease at night at 0.01 confidence level.The distinct behavior of OA during days with episodes and no episodes is illustrated in the box plots of Fig. 10.The box limits are the 25 and 75 percentiles, the black line is the median, the black dot corresponds to the mean and the whiskers represent the standard deviation.HOA and BBOA are characterized by having a larger variability at night, because the box limits and the error bars are larger, denoting that episodes differ from each other.The results in Fig. 10 indicate that primary organic aerosol (POA) is responsible for the buildup of pollution at night during episodes.These results are also consistent with chemical speciation performed by Villalobos et al., 2015 in Santiago that found that during winter, 58% of the organic carbon was generated by wood smoke.It is interesting to note that during episodes the wind speed is lower and air masses stagnate over the city, although these conditions seem to favor aging of the organic aerosol, this is not what was observed.A recent study in London (Young et al., 2015) also found that there is an increase at night during cold periods of the organic aerosol associated to biomass burning.These results are consistent with use of different types of biofuels (coal, wood, peat, biogas,  etc.) for space heating and its ubiquity across many cities of the world may have important implications for global warming models.

CONCLUSIONS
The winter campaign was carried out in downtown Santiago de Chile revealed that during days in which pollution reached extreme levels, PM 2.5 had a different temporal behavior than days with lower pollution.In days in which the average PM 2.5 was higher than 50 µg m -3 , the concentration at night of several pollutants was much higher than the concentration during the day.The average of all measurements during episodes, (between 6 am-7 pm) divided by the same average without episodes, is lower than the average (between 8 pm-5 am) during episodes divided by the same average without episodes.This ratio is much larger for NO, CO and BC, i. e. primary pollutants.For secondary pollutants (NH 4 , NO 3 , SO 4 , NO 2 ), a small increase or no change was observed at night during episodes.This is an indication that during episodes the primary emissions are responsible for the high PM 2.5 levels observed.
Consequently, secondary pollutants play a minor role during episode days in this part of the city.When the organic aerosol is separated into hydrocarbon-like organic aerosol (HOA), biomass burning organic aerosol (BBOA) and oxygenated organic aerosol, the same conclusions are reached, i.e., the newer aerosols (HOA and BBOA) are predominant at night during episodes.This behavior allows identifying biomass burning and vehicle emissions as the sources that are responsible for the high levels reached during episodes in Santiago.The meteorological conditions that trigger these high pollution episodes (low wind speed and temperature inversion) are common in many cities that suffer air contamination, thus techniques similar to the one used in this work can be applied to determine the sources of PM 2.5 .The results in Santiago and many cities of the world indicate that during winter, there is an enhancement of light absorption because an increase in carbonaceous aerosol emissions.Identification of the percentage of BC and BrC as well as the season and time of its emissions might be important to improve global warming model predictions (Liu et al., 2015).

Fig. 1 .
Fig. 1.Map of Santiago de Chile showing the location of the sampling sites.The arrow indicates the predominant wind direction during the afternoon.

Fig. 2 .
Fig. 2. The top curve shows PM 2.5 at Parque O'Higgins site measured with a beta monitor.The bottom curve shows the time series of the combined PM 1.0 ACSM measurements (organics + nitrate + sulfate + ammonium + chloride) and black carbon at Usach site.Correlation between the curves is R 2 = 0.5.

Fig. 5 .
Fig. 5.Diurnal profile of several pollutants and meteorological parameters, separated for days with and without episodes.Horizontal scale shows the time of day.

Fig. 6 .
Fig. 6.Mass spectra of the six factors from the PMF calculation.

Fig. 9 .
Fig. 9.Diurnal profile of the organic components, separated for days with and without episodes for the measurement period.

Fig. 10 .
Fig. 10.(a) Box plots of the PM 1.0 components for the day and night period during episodes, (b) PM 1.0 components during days without episodes.The black dot is the mean of each series.

Table 1 .
Composition of PM 2.5 and PM 1.0 in Santiago.In 2005 and 2003 measurements of OC were performed with thermo-optical methods.In 2011 and 2012 organic components were measured with the ACSM.

Table 2 .
Average of all measurements between 6 am-7 pm. with episodes divided by the same average without episodes.Second column, days with episodes between 8 pm-5 am divided by days without episodes.

Table 3 .
Average of the meteorological variables for the measurement period (March-June 2012).