Chemical Apportionment of Southern African Aerosol Mass and Optical Depth

This study characterizes the aerosol over extrat-ropical and tropical southern Africa during the biomass burning season by presenting an aerosol mass apportionment and aerosol optical properties. Carbonaceous aerosol species account for 54% and 83% of the extratropical and tropical aerosol mass, respectively, which is consistent with the fact that the major source of particulate matter in southern Africa is biomass burning. This mass apportionment implies that carbonaceous species in the form of organic carbon (OC) and black carbon (BC) play a critical role in the aerosol optical properties. By combining the in situ measurements of aerosol mass concentrations with concurrent measurements of aerosol optical properties at a wavelength of 550 nm, it is shown that 80–90% of the aerosol scattering is due to carbonaceous aerosol, and the derived mass scattering cross sections (MSC) for OC and BC are 3.9±0.6 m 2 /g and 1.6±0.2 m 2 /g, respectively. Derived values of mass absorption cross sections (MAC) for OC and BC are 0.7±0.6 m 2 /g and 8.2±1.1 m 2 /g, respectively. The values of MAC imply that ∼26% of the aerosol absorption in southern Africa is due to OC, with the remainder due to BC. The results in this study provide important constraints for aerosol properties in a region dominated by biomass burning and should be integrated into climate models to improve aerosol simulations.


Introduction
Biomass burning is a major source of particulate matter in the atmosphere over Africa during the dry season.In southern Africa alone, biomass burning consistently accounts for 25% of the total number of fires that occur every year on the Correspondence to: B. I. Magi (brian.magi@noaa.gov)planet (Giglio et al., 2003(Giglio et al., , 2006)), but in situ measurements of tropospheric aerosol mass concentrations in that region are rare.This was part of the motivation for the Southern African Research Initiative field campaign in August and September 2000 (SAFARI-2000), and resulted in a number of publications from about 2002 to present, many of which are summarized by Swap et al. (2003).
This study synthesizes in situ measurements of aerosol mass concentrations obtained by aircraft in the southern African regional haze and originally published in SAFARI-2000 studies by Eatough et al. (2003), Gao et al. (2003), and Kirchstetter et al. (2003).Measurements of aerosol mass concentrations made in smoke plumes are not discussed here, but there are detailed discussions by Eatough et al. (2003), Gao et al. (2003), and Kirchstetter et al. (2003).The synthesis of the regional haze measurements requires more than reviewing the research since there was a technical problem with part of the aerosol sampling system referred to by Gao et al. (2003) and by Kirchstetter et al. (2003).A solution to the problem is described in Sect. 2. Without this solution, aerosol mass concentrations presented by Eatough et al. (2003), Gao et al. (2003), and Kirchstetter et al. (2003) could only be analyzed independently.Hence, the synthesis provides a unified measurement-based estimate of aerosol mass concentrations over a region of the world that is poorly simulated in most transport and circulation models (e.g.Ginoux et al., 2006;Kinne et al., 2006;Matichuk et al., 2007;Magi et al., 2009).
The aerosol mass concentrations in this synthesis are combined with self-consistent aerosol optical properties discussed by Magi et al. (2008) to derive values of the mass scattering cross section (MSC) and the mass absorption cross sections (MAC) at a wavelength of 550 nm for measured aerosol species that are relevant to models.Finally, the total scattering and absorption of the southern African aerosol is apportioned into species contributions.Methods Published by Copernicus Publications on behalf of the European Geosciences Union.
B. I. Magi: Chemical apportionment of southern African aerosol mass are discussed in Sect.2, results and analysis are presented in Sect. 3 and Sect. 4, and conclusions are summarized in Sect. 5. Gao et al. (2003) and Kirchstetter et al. (2003) discussed the methods of data collection and analysis, but they are briefly reiterated here.The teflon filter samples described by Gao et al. (2003) and the quartz filter samples described by Kirchstetter et al. (2003) (hereafter, referred to as G2003 and K2003, respectively) were collected in tandem on a manifold that pulled air through an inlet from outside the aircraft, across the filters, and exited the aircraft through an outlet.The inlet (Sinha et al., 2003) had a 2-3 µm particle diameter cut-off such that all reported results apply to PM 2.5 (i.e.particulate matter, PM, with diameters less than 2.5 µm).Independent in situ evidence shows that the number concentration of the southern African aerosol in the polluted layer (below about 5 km) is dominated by particles with diameters less than 1.0 µm (Li et al., 2003;Posfai et al., 2003;Formenti et al., 2003).

Background
Teflon filter samples were analyzed by Gas Chromatography-Mass Spectrometry (GC-MS) for PM 2.5 aerosol mass concentration, and the mass concentrations of PM 2.5 ionic species like sulfate (SO = 4 ) and nitrate (NO − 3 ), among others.For simplicity in this study, ionic species are referred to as SO 4 and NO 3 .Quartz filter samples were analyzed by Evolved Gas Analysis (EGA) for PM 2.5 carbonaceous aerosol mass concentration, which was usually apportioned into organic carbon (OC) and black carbon (BC) components.G2003 and K2003 provide much more detailed descriptions about the GC-MS and EGA techniques, respectively.
The sampling system used to determine aerosol mass concentrations reported by Eatough et al. (2003) (hereafter, referred to as E2003) was completely independent of the manifold used to collect samples described by G2003 and K2003, but both sampling systems were on the same aircraft (see Appendix in Sinha et al., 2003).Using the Particle Concentrator Brigham Young University Organic Sampling System (PC-BOSS) to collect filter samples, E2003 used Temperature Programmed Volatization Analysis (TPVA), EGA, and Ion Chromatographic Analysis (ICA) to report mass concentrations of OC, BC, SO 4 , NO 3 , and semi-volatile organic carbon (SVOC).No other study has reported SVOC in southern Africa, and SVOC is generally thought to be lost during in situ sampling.OC and BC as reported by E2003 and K2003 does not include this additional mass.
Samples reported by E2003 were collected at a lower frequency than samples discussed by G2003 and K2003, but on average the samples all characterized the same aerosol.Figure 1 shows the locations of samples discussed by E2003, G2003, and K2003.Formenti et al. (2003) discussed samples collected during SAFARI-2000, but further west of the major burning regions.Ground-based results reported by Piketh et al. (1999) and Maenhaut et al. (1996) were collected in southern to eastern South Africa, but our analysis is restricted to aerosol samples collected in the troposphere since groundbased filter data are not always representative of the aerosol characteristics aloft.
The results in this study are partitioned into categories based on the parcel back-trajectory analysis by Magi and Hobbs (2003).Namely, samples collected when backtrajectories were from the south or east are one category, while those collected when back-trajectories were from the north or west are a second category.The sample area roughly covers northeast South Africa, eastern Botswana, and southern Zambia (Fig. 1), and therefore parcels arriving from the south or east are extratropical (south of 22.5 • S) in origin, while those from the north or west are tropical (north of 22.5 • S) in origin.Therefore, samples are referred to as either characterizing the "extratropics" or "tropics" based on this discussion.
Since the tropics of southern Africa are more impacted by fires (Fig. 1), and the extratropics are relatively more impacted by industrial and dust sources (Piketh et al., 1999;Bond et al., 2007), differences in aerosol properties (such as chemical composition, size, or number concentration) between samples characteristic of the tropics and the extratropics are expected.The evidence showing the aerosol characteristics is summarized in Table 1, where the median aerosol optical properties at a wavelength of 550 nm and median aerosol physical properties are listed, along with the respective interquartile range.Aerosol optical depth (AOD) is one of the most commonly reported aerosol optical properties, and is calculated as the vertical integral of aerosol extinction (EXT) or retrieved from remote sensing measurements.EXT itself is the sum of aerosol scattering (SCA) and absorption (ABS), where EXT, SCA, and ABS are in units of m −1 or Mm −1 .Single scattering albedo (SSA) is defined as SCA/EXT and physically represents the probability that radiation incident on a particle is scattered rather than absorbed.
The summary statistics in Table 1 are based on data discussed by Magi et al. (2008) and by Magi (2006).Over the tropics, AOD is significantly higher and SSA is significantly lower.AOD is higher due to the factor of two increase in accumulation mode particle concentration (N 2 in Table 1), and very little of the difference can be attributed to changes in the shape of the size distribution (α) or an increase in coarse mode particle volume concentration (C/F).Since relative humidity (RH) affects ambient aerosol properties by increasing the size of the particles, which subsequently increases the AOD (e.g.Hegg et al., 1993), the median RH is listed in Table 1 as well.Magi and Hobbs (2003) showed that aerosol scattering coefficients in southern Africa are not significantly affected when RH is less than ∼50%, so it is clear from Table 1 that RH cannot explain the AOD differences between the tropics and extratropics.The lower SSA in the tropics is due to a change in the overall chemical composition of the aerosol.The evidence supporting this assertion and the chemical species that contributed to the increase in AOD in the tropics is discussed in Sect.3.

Miscalibrated flow meters
Mass concentrations derived from filters depends on the integrated flow volume (m 3 ) that passes across the filters.The three flow meters (Hastings Model TNALL) used to measure the flows for G2003 and K2003 were calibrated against a bubble flow meter before and after the SAFARI-2000 campaign and compared to a separate flow meter on the aircraft during the campaign.Each flow meter had a different set of calibration coefficients.In all cases, total flow was reported at standard temperature and pressure (STP, 273 K, 1013 hPa) so that mass concentrations derived at different altitudes could be directly compared.
After the SAFARI-2000 campaign, we realized that the calibration coefficients of the flow meters were dependent on whether multiple flow meters were in use.The end result was that the total flows measured by the flow meters used for the filters reported by G2003 were in error by some unknown value.The flow meter used with the filters reported by K2003 were not affected because of the position of the flow meter on the aircraft manifold.This was verified qualitatively (simple observations during SAFARI-2000), but the flow meters were removed from the manifold before the calibration coefficients for the flow meters used for filters described by G2003 could be quantitatively verified.Therefore, although there were tandem filter samples, absolute comparisons between results by G2003 and by K2003 were not possible since one of the filter measurements (usually the teflon filters) used a flow meter with an unknown calibration constant.
As evidence of the problem introduced by the miscalibrated flow meter, the mass concentrations reported by G2003, K2003, and E2003 are summarized in Table 2. OC and BC reported by K2003 generally agreed with results by E2003 to within the bounds of the interquartile range or within the ∼30% measurement uncertainty.The agreement is better in the extratropics (Table 2a) than in the tropics (Table 2b), but this could be because the sample size of E2003 filters representative of the tropics is only four.However, results by G2003 (measured at the same time as K2003) do not agree with the results by E2003 (referring to PM 2.5 , SO 4 , and NO 3 in Table 2).The discrepancy is most clear when the ratio of aerosol mass concentrations from the tropics to the extratropics is examined.G2003 results suggest a tropics to extratropics PM 2.5 ratio of 0.8, while E2003 results suggest a ratio of 3.1.The corresponding ratio of OC for K2003 3.5, while for E2003 is 5.8.Thus, G2003 data suggests that PM 2.5 decreases in the tropics, while E2003 suggests an increase.Yet, OC reported by K2003 and E2003 clearly increases in the tropics due to more widespread burning as observed by www.atmos-chem-phys.net/9/7643/2009/Atmos.Chem.Phys., 9, 7643-7655, 2009 Table 2.
Comparison of median aerosol mass concentrations (µg/m 3 ) and interquartile range (IQR) reported by Gao et al. (2003), G2003, and Kirchstetter et al. (2003), K2003, versus those reported by Eatough et al. (2003), E2003, for samples representative of Southern Hemisphere African (a) extratropics and (b) tropics.K2003 had a reasonable calibration, while G2003 did not, and these results have not been corrected to account for the miscalibrated flow meters (Sect.2).E2003 results were based on measurements from an independent inlet.satellite (Fig. 1).Table 1 provides further evidence of the discrepancy by showing a significant increase in AOD and accumulation mode aerosol number concentrations (Sect.2.1) with no evidence supporting a significant change in the dominant particle size of the aerosol size distribution (i.e.C/F and α in Table 1).The major reason for the difference in AOD in the samples is thus due to an increase in aerosol number concentration, and therefore an increase in PM 2.5 mass is expected.G2003 results in Table 2 contradict this evidence, while E2003 and K2003 support the evidence.

Proxy calibration
In order to partition the aerosol mass or make any generalizations about measured aerosol mass, absolute mass concentrations must be used.A working solution to overcome the problem of miscalibrated flow meters was developed based on three separate pieces of information: 1. we knew which flow meters were used to measure flow for each sample, 2. on four occasions, duplicate samples were collected by sampling aerosol on pairs of the same type of filters (twice with teflon, twice with quartz) and monitoring the flow rate through each filter using different flow meters, and 3. the independent sampling system (and flow meter) described by E2003 was on the same aircraft.The simultaneous filter samples, which should report nearly identical mass concentrations, were crucial to formulating a working solution.The intended purpose of the simultaneous filter samples was to characterize the noise inherent in the measurements of filter mass for the techniques described by G2003 and by K2003, but they resulted in a convenient solution to the miscalibrated flow meter.The results of the simultaneous filter samples are listed in Table 3 as Test 1 and Test 2 (obtained on two different days), and it is clear that there are discrepancies well beyond the expected noise (which should have been much less than 10%) in reported mass concentration.Namely, the Test 1 results from "Flow Meter 1" for teflon filters (PM 2.5 , NO 3 , SO 4 ) and for quartz filters (OC, BC) should be roughly the same as the results from "Flow Meter 2".Similarly, Test 2 results from "Flow Meter 2" and "Flow Meter 3" should be roughly the same as well.Generally, Flow Meter 2 was used to derive measurements reported by K2003, while Flow Meter 1 and Flow Meter 3 were used to derived measurements from the teflon filters.
Purely by coincidence, the teflon filters obtained from the tropics were derived using Flow Meter 1 while the extratropical samples were obtained using Flow Meter 3.All quartz filters were obtained using Flow Meter 2, which was correctly calibrated (i.e.unaffected when multiple flow meters were in use).Using Table 3, the median ratio of mass concentrations reported by Flow Meter 2 to mass concentrations reported by Flow Meter 1 was 3.4 (interquartile range of 3.1 to 4.3), while the ratio of Flow Meter 2 to Flow Meter 3 was 0.5 (0.4 to 0.5).Assuming the mass (in µg) reported in Table 3 should be equal, the fact that the ratios were greater than or less than unity is due entirely to difference in the flow (in m 3 ) and we can use this as a proxy for direct calibration.
Evidence supporting this proxy calibration using Table 3 comes from the third piece of information.Because all the teflon filter samples (i.e.G2003) from the tropics were collected using Flow Meter 1, the ratio of mass concentrations reported by E2003 for the tropics to mass concentrations reported by G2003 for the tropics (listed in Table 2) should be similar to the ratio of Flow Meter 2 to Flow Meter 1 (stated above as 3.4).The ratio of mass concentrations in the tropics is 3.8 (interquartile range of 3.2 to 6.1), which encompasses the median value of the ratio of Flow Meter 2 to Flow Meter 1 of 3.4 reported above.Similarly, the ratio of E2003 to G2003 in the extratropics is 0.5 (0.5 to 0.7), and encompasses the median value of 0.5.
Using the information described above as a proxy for calibration coefficients for the miscalibrated flow meters (i.e.Flow Meter 1 and Flow Meter 3), the mass concentrations derived from the teflon filters (i.e.G2003) were corrected.The statistics from the corrected mass concentrations are listed in Table 4, noting that values of OC and BC are nearly identical to those listed in Table 2 since the carbon aerosol mass concentrations were mainly derived using the properly calibrated flow meter (Flow Meter 2).PM 2.5 , SO 4 , and NO 3 from G2003 agree to within 23% with E2003 after we apply the proxy calibration correction, which is a 44% improvement over the pre-correction agreement.Equally as significant is the fact that the tropics to extratropics ratio of corrected mass concentrations (see end of Sect.2.2) in Table 4 now agree to within 1% (a 57% improvement over pre-corrected  4 affect some of the published results from past studies.K2003 results are not affected, and in fact their Table 1 includes the individual mass concentrations of OC and BC used in this study.G2003 presented results based on the relative contribution of SO 4 and NO 3 to PM 2.5 , but these numbers are consistent with those in this study since the same correction factor was applied to SO 4 , NO 3 , and PM 2.5 .Table 2 in G2003 lists mass concentrations from "Background Aerosol Samples", and Table 2 in Sinha et al. (2003b) lists mass concentrations for particulate matter from various "sectors".Coincidentally and conveniently, Table 2 in this study is consistent with Table 2 in both G2003 and Sinha et al. (2003b).Mass concentrations should be corrected to better match Table 4 in this study.Some of the various statements by Sinha et al. (2003b) would need to be revised to reflect the results of the correction, but most of their conclusions remain valid.Other studies do not appear to be affected.

PM 2.5 aerosol mass
The statistics of the speciated mass concentrations from all the filter samples (teflon, quartz, and those reported by E2003) are listed in Table 5. PM 2.5 mass concentrations are about 12 µg/m 3 in the extratropics and increase by a factor of four in the tropics to nearly 48 µg/m 3 .This increase is consistent with the data in Fig. 1 which shows that the tropics are where most of the burning occurs in August and September and with aerosol physical and optical properties listed in Table 1.The speciated mass apportionment reveals that the primary driving force for the increase in PM 2.5 is an increase in OC and BC mass.NO 3 mass increases by a factor of 14 Comparison of median aerosol mass concentrations (µg/m 3 ) and interquartile range (IQR) reported by Gao et al. (2003), G2003, and Kirchstetter et al. (2003), K2003, versus those reported by Eatough et al. (2003), E2003, for samples representative of Southern Hemisphere African (a) extratropics and (b) tropics after we applied corrections based on Table 3 to a concentration of about 2.6 µg/m 3 .Although this NO 3 mass concentration is greater than typical background concentrations reported by Sinha et al. (2003a) and Tripathi et al. (1996), NO 3 still accounts for less than 5% of the PM 2.5 mass.
Both SO 4 and SVOC aerosol mass concentrations in Table 5 are not significantly different in the tropics compared to the extratropics.The similarity in SO 4 could partly be attributed to additional SO 4 sources in the extratropics that are not biomass burning in origin (Piketh et al., 1999;Bond et al., 2007).SVOC mass statistics are based on ten samples in the extratropics and four samples in the tropics, so the limited sample sizes may not be sufficient to support any conclusion about SVOC in the tropics versus the extratropics.The interquartile range of SVOC, for example, is much larger in the tropics, presumably indicating that the fire and/or biomass characteristics also were different among the four samples.The only conclusion in terms of SVOC that is offered in this study is exactly what E2003 pointed out: most filter samples do not include a measure of SVOC and therefore, based on Table 5, PM 2.5 reported by G2003 and K2003 (and others using similar analysis techniques) may be underestimated by 11-33%.

PM 2.5 mass apportionment
Using the information in Table 5, and converting the measured species to common molecular forms, the PM 2.5 aerosol mass is apportioned into species contributions.The median values of the corrected mass concentrations of potassium (K + ) by G2003 are 0.15 and 0.65 µg/m 3 for the extratropics and tropics, respectively.This suggests that 94% (extratropics) and 86% (tropics) of the SO 4 and NO 3 ions are not www.atmos-chem-phys.net/9/7643/2009/Atmos.Chem.Phys., 9, 7643-7655, 2009 balanced by K + , and it is assumed for simplicity that they are entirely neutralized by ammonium (NH 4 + ).The ionic aerosol species (SO 4 and NO 3 ) are converted to ammonium sulfate ((NH 4 ) 2 SO 4 ) and ammonium nitrate (NH 4 NO 3 ) by multiplying the mass of SO 4 and NO 3 by 1.38 and 1.29, respectively.
Conversion of OC to organic matter (OM) to account for the non-carbon mass of molecular forms of OC is subject to considerable uncertainty.Dentener et al. (2006) suggest a conversion factor of 1.4.Turpin and Lim (2001) suggest that 1.4 should be considered a lower bound to estimates and that a value of 1.6±0.2 and 2.1±0.2 are more representative of urban and non-urban aerosols, respectively, but they point out that even this range is subject to re-examination considering how little we still understand about the many compounds that together comprise "OC" (e.g.Kanakidou et al., 2005).A OM/OC ratio for the extratropics and tropics is derived by assuming that 100% of the PM 2.5 mass can be attributed to NH 4 NO 3 , (NH 4 ) 2 SO 4 , BC, and OM (including SVOC makes no difference in this case).Thus, OM/OC = (PM 2.5 -NH 4 NO 3 -(NH 4 ) 2 SO 4 -BC)/OC.This gives a value of 1.4 (interquartile range is 1.0 to 1.6) for OM/OC in the extratropics and 2.1 (1.6 to 3.0) for OM/OC in the tropics.The median values are within the broad range of 1.4-2.3suggested above.
The chemical apportionment of PM 2.5 mass is shown in Fig. 2 In the extratropics, 54% of the aerosol mass is OM and BC, while the remaining 46% is comprised of NH 4 NO 3 and (NH 4 ) 2 SO 4 .As discussed above, (NH 4 ) 2 SO 4 is roughly the same in the tropics and the extratropics, but the dramatic increase in OM and BC in the tropics results in an aerosol comprised of 83% carbonaceous components.If SVOC is included, the contribution of carbonaceous particles to PM 2.5 mass increases to 69% in the extratropics and 85% in the tropics.

Aerosol optical depth apportionment
In this Section, the apportionment of AOD at a wavelength of 550 nm is described.Most of the speciated aerosol mass concentrations that comprise the median values listed in Table 5 were coincident with measurements of SCA and ABS (with a PM 2.5 aerosol cut-off) at a "dry" relative humidity (RH) of 30% (see Magi et al., 2003Magi et al., , 2008 for details about the measurements of SCA and ABS).Aerosol mass concentrations in this study are reported at standard temperature and pressure (STP, 273 K, 1013 hPa) and for dry aerosol mass, so the dry SCA and ABS are also adjusted to STP.Thus, the species mass concentrations can be related to SCA and ABS by assuming that each species plays independent roles (i.e. an external mixture) in determining the totals by where M i is the mass concentration of NH 4 NO 3 , (NH 4 ) 2 SO 4 , OM, and BC and MSC i and MAC i are the mass scattering and mass absorption cross sections (e.g.Bond and Bergstrom, 2006), in units of m 2 /g.SCA and ABS in these equations and those that follow are at a dry RH and at STP.All the aerosol species contribute to scattering.For the multiple linear regression, the mass concentrations of NH 4 NO 3 and (NH 4 ) 2 SO 4 are combined together and OM and BC are also combined together, such that we modify Eq. ( 1) to be where M ionic is the sum of NH 4 NO 3 and (NH 4 ) 2 SO 4 and M carbon is the sum of OM and BC.The regression is unstable if the four aerosol species are separately considered, mainly due to the relatively small sample size.Since all the scattering is attributed to these aerosol species, it is also important to consider the molecular forms of the species (i.e.NH 4 NO 3 , (NH 4 ) 2 SO 4 , and OM as opposed to NO 3 , SO 4 , and OC).The values of M ionic and M carbon are then partitioned using Mie theory (Bohren and Huffman, 1983;Mätzler, 2002;Bond et al., 2006; http://www.hiwater.org/)approximations to derive MSC OC , MSC BC , MSC SO 4 , and MSC NO 3 .Equation ( 2) can be simplified since NH 4 NO 3 and (NH 4 ) 2 SO 4 do not significantly absorb visible radiation (e.g.Tang and Munkelwitz, 1994), while BC and OC are both potential absorbers (e.g.Bond and Bergstrom, 2006;Kirchstetter et al., 2004).Therefore Eq. ( 2) becomes   noting that BC is the major absorbing aerosol species in the mid-visible, while OC absorption seems to vary depending on the location.Kirchstetter et al. (2004) showed that OC accounts for about 7-24% of the total absorption at a wavelength of 550 nm, with an imaginary refractive index of about 0.03.Kanakidou et al. (2005) summarized research about OC and suggested an imaginary refractive index for OC of about 0.006.Transport models discussed by Kinne et al. (2006) assumed an imaginary refractive index of 0.03, similar to Kirchstetter et al. (2004).Even a very small OC contribution to total absorption is important, however, since OC mass burden is often much greater than BC mass burden (e.g.Bond et al., 2004).
The values for MSC and MAC, determined by solving Eqs. ( 3) and (4) using multiple linear regression, are listed in Table 6 along with the standard errors.Also listed in Table 6 are MSC and MAC calculated from Mie theory (Bohren and Huffman, 1983;Mätzler, 2002;Bond et al., 2006; http://www.hiwater.org/),where the assumptions used to calculate the theoretical MSC and MAC are listed in Table 7 (lognormal size distribution diameter and standard deviation, density, and refractive indices at 550 nm).The values in Table 7 for OC are similar to those used in models (e.g.Kinne et al., 2006) and suggested in the review study by Kanakidou et al. (2005), while the density of OC is based on Turpin and Lim (2001).The values for BC are based on Kirchstetter et al. (2004) and Bond and Bergstrom (2006).The values we used to simulate NH 4 NO 3 and (NH 4 ) 2 SO 4 are based on Tang and Munkelwitz (1994).
Values of MSC OC , MSC BC , MSC SO 4 , and MSC NO 3 are derived by partitioning the values of MSC ionic and MSC carbon from Eq. ( 3) according to the theoretical contributions predicted by Mie theory.Thus MSC SO 4 =0.47*1.38*MSCionic and MSC NO 3 =0.53*1.29*MSCionic , where the 0.47 and 0.53 are the fractional contributions to MSC ionic and 1.38 and 1.29 are the molecular conversion factors.From MSC carbon , MSC OC =0.59*1.7*MSCcarbon and MSC BC =0.41*MSC carbon , where 0.59 and 0.41 are the fractional contributions to MSC carbon and 1.7 is the mean ratio of OM to OC.In all cases, using literature-based assumptions to simulate the aerosol components, the measurement-based MSC and MAC agree with theoretical values of MSC and MAC.These results are analyzed in the next Section.

Aerosol mass
The evidence presented above clearly shows that the more concentrated burning in tropical southern Africa results in an aerosol that is significantly different than extratropical southern Africa.These differences apply to aerosol optical properties and number concentrations (Table 1), PM 2.5 mass concentrations (Table 5), and even PM 2.5 mass apportionment (Fig. 2).The major differences are that tropical southern African aerosol is composed of 83% carbonaceous aerosol, www.atmos-chem-phys.net/9/7643/2009/Atmos.Chem.Phys., 9, 7643-7655, 2009 while the extratropical southern African aerosol is composed of 54% carbonaceous aerosol and PM 2.5 mass increases by a factor of four in the tropical aerosol due mainly to increases in OC mass.The extratropical aerosol samples discussed in this study are most likely biased towards the more intense biomass burning since the sample locations (Fig. 1) are located east of the Namib and Kalahari deserts.Sulfate aerosol contributions from industrial sources in northeast South Africa (e.g.Bond et al., 2007;Piketh et al., 1999) may also affect the mass apportionment we showed in Fig. 2a.The fractional contributions to overall PM 2.5 mass should be considered when simulating the aerosol over southern Africa.Formenti et al. (2003) characterized southern African aerosol properties near Namibia (15-25 • S, 10-15 • E) during SAFARI-2000.Using their reported values of NH 4 NO 3 , (NH 4 ) 2 SO 4 , OM, and BC mass concentrations, carbonaceous aerosol makes up about 70-90% of PM 2.5 mass.Even considering all of the reported species in Formenti et al. (2003), PM 2.5 aerosol is composed of about 60-75% carbonaceous aerosol.Although the samples discussed by Formenti et al. (2003) are west and south of major burning regions, particles emitted from biomass burning are often transported from both tropical and extratropical burning regions over Namibia and over the southeast Atlantic Ocean (e.g.Garstang et al., 1996).Formenti et al. (2003) also confirmed this transport pathway with an analysis of backtrajectories of their sample locations.
The mass concentrations of NO 3 , SO 4 , OC, and BC reported by Formenti et al. (2003) range from 1.0-1.4,1.9-2.0,6-20, and 1.0-2.0µg/m 3 , respectively.Values of NO 3 , OC, and BC in Table 5 are all very similar, while SO 4 mass concentration is a factor of two larger.The higher SO 4 concentration is expected since many of samples in this study were affected by proximity to concentrated burning (tropics) or proximity to industrial sources (extratropics).Ruellan et al. (1999) collected aircraft-based in situ aerosol samples during the Experiment for Regional Sources and Sinks of Oxidants (EXPRESSO) field campaign in November and December 1996 over Northern Hemisphere equatorial Africa (3-5 • N, 17-19 • E).Fires in this region in November and December are as numerous as those in tropical southern Africa during August and September, so there should be some similarities in the aerosol regardless of the differences in burning characteristics (i.e.vegetation, combustion completeness).Ruellan et al. (1999) reported mean OC and BC mass concentrations of about 22±11 and 5±2 µg/m 3 , respectively, between the surface and about 4 km.This is very similar to median OC and BC mass concentrations of 17 µg/m 3 (12-23 µg/m 3 interquartile range) and 3.4 µg/m 3 (1.8-4.5 µg/m 3 interquartile range), respectively, reported in Table 5 for tropical southern Africa.
Across the entire African region, there is variability in aerosol mass concentration in the polluted layer, but the magnitudes generally agree in terms of carbonaceous aerosol.Based on a small number of studies from EXPRESSO and SAFARI-2000, the carbonaceous aerosol mass concentration seems to decrease from north to south and from east to west.This is consistent with fire occurrence patterns derived from MODIS (Giglio et al., 2003(Giglio et al., , 2006)).Both this study, which is based on data originally described by E2003, G2003, and K2003, and the study by Formenti et al. (2003), suggest that at least half the PM 2.5 aerosol mass concentration during the biomass burning season is composed of carbonaceous aerosol.This fraction increases significantly as PM 2.5 mass increases.
The correlation of OC, BC, NO 3 , and SO 4 mass with PM 2.5 mass is shown in Fig. 3 and, with the exception of SO 4 , all are significantly correlated (p values less than 0.05, implying the correlation is significant with greater than 95% confidence).OC is clearly the main factor driving increases in PM 2.5 mass, with r 2 =0.87.SO 4 on the other hand is not significantly correlated, and only about 7% of the variance in PM 2.5 can be explained by SO 4 .
The significant correlations of OC and BC mass with PM 2.5 mass is not restricted to the tropics or the extratropics.The correlation of NO 3 and SO 4 with PM 2.5 , however, does depend on the regional subset.Namely, SO 4 is significantly correlated with PM 2.5 in the extratropics (r 2 =0.67), and NO 3 is significantly correlated with PM 2.5 in the tropics (r 2 =0.60), while the opposite in both cases is not true.The better correlation of NO 3 with PM 2.5 in the tropics, where burning is more prevalent, agrees with Hobbs et al. (2003), who suggested NO 3 enhancements in aged smoke due to secondary (photochemical) formation.The evidence presented in this study suggests that the major SO 4 sources in the extratropics are split between biomass burning and fossil fuel burning in northeastern South Africa and that this explains why SO 4 is better correlated with PM 2.5 in the extratropics than the tropics.

Mass scattering and mass absorption cross sections
The unifying characteristics of the tropical and extratropical southern Africa aerosol are the values of MSC and MAC, especially with respect to carbon species (OC and BC).Using the multiple linear regression technique described in Sect.3.3, MSC OC =3.9±0.6 m 2 /g and MSC BC =1.6±0.2 m 2 /g (Table 6).Based on previous work, all the absorption was assumed to be attributed to OC and BC, and the multiple linear regression results suggested MAC OC =0.7±0.6 m 2 /g and MAC BC =8.2±1.1 m 2 /g (Table 6).MSC SO 4 and MSC NO 3 are both very small and the uncertainty in the multiple linear regression coefficients is large.All of the measurement-based results in this study, however, are consistent with Mie theory, although the theoretical values are themselves dependent on assumptions listed in Table 7.
In a detailed review of past research related to BC absorption and a critical analysis of BC absorption by Bond and Bergstrom (2006), they suggested that MSC BC varied from about 1.6 to 3.7 m 2 /g and MAC BC =7.5±1.2 m 2 /g.Values of MSC BC and MAC BC listed in Table 6 agree to within uncertainty with Bond and Bergstrom (2006).OC optical properties, particularly absorption, are a much more poorly understood aspect of aerosol science.Kirchstetter et al. (2004) derived a value of MAC OC =0.6 m 2 /g from smoke samples collected during SAFARI-2000, while Sun et al. (2007) suggest MAC OC =1.0 m 2 /g based on a theoretical model of a combustion aerosol.The value of 0.7±0.6 m 2 /g in this study agrees well with these past studies, especially considering the much different methodologies.The value of MSC OC listed in Table 6 is on the high end of previously published values of MSC OC , which range from 1-5 m 2 /g and are sometimes based on data, and sometimes based on assumptions (e.g.White, 1990;Liousse et al., 1996;Magi et al., 2009).
The single scattering albedo (SSA) at 550 nm is defined as MSC/MEC where MEC is the mass extinction cross section and is the sum of MSC and MAC.MSC/MEC and SCA/EXT as described in Sect.2.1 are equivalent ways of calculating SSA.SSA SO 4 and SSA NO 3 are therefore equal to 1 in this study, but SSA OC =0.85 and SSA BC =0.17.Bond and Bergstrom (2006) suggest that SSA BC =0.2-0.3,so the derived value in this study is at the low end of that range.In a modeling study, Magi et al. (2009) suggested SSA OC =0.97 and SSA BC =0.21, but also showed that the overall aerosol SSA simulated by the model was biased high compared to measurements.The high biases in modeled SSA (e.g.Kinne et al., 2006) would improve if models simulated a more strongly absorbing OC component in regions of the world impacted by biomass burning, as the data presented here suggests.

Apportionment of scattering and absorption
Using the MSC and MAC derived from multiple linear regression, the scattering and absorption in the extratropics and the tropics is apportioned.In the apportionment, the molecular forms of the species are used for scattering since all the aerosol mass is responsible for scattering solar radiation.For   absorption, OC and BC are used since there is no evidence suggesting that OM is more absorbing than OC.The aerosol optical properties apportionment for the extratropics is shown in Fig. 4 while the tropical apportionment is shown in Fig. 5.Both figures reveal similar trends, but NH 4 NO 3 and (NH 4 ) 2 SO 4 account for about twice as much scattering in the extratropics than in the tropics.Carbonaceous aerosol accounts for an overwhelming 80-90% of the total scattering.
The evidence presented in Figs.4b and 5b suggests that OC accounts for about 26-27% of the absorption, while BC accounts for the remainder.These numbers agree with SAFARI-2000 results by Kirchstetter et al. (2004), who used a much different methodology and showed that OC accounts for 7-24% of the absorption at 550 nm for freshly emitted smoke.Although the differences between Kirchstetter et al. (2004) and this study are not statistically significant, the generally higher contribution of OC to absorption in this study might be due to enhanced OC absorption from secondary particle formation as the smoke ages in the atmosphere.In terms of the scales involved in simulating an aerosol in climate models, it is perhaps more appropriate to model the properties of an aged aerosol.

Conclusions
Using synthesized results from three previously published studies, the southern African aerosol during the biomass burning season was characterized.The synthesis provided speciated aerosol mass representative of tropical and extratropical southern Africa.PM 2.5 mass in the tropics is a factor of four greater than in the extratropics (Table 5), and this is driven primarily by increases in carbonaceous aerosol mass from the more widespread occurrence of fires in the tropics.The magnitude and species contributions to total aerosol mass in this study (Fig. 2), after applying corrections described in Sect.2, agree well with other studies of African aerosols dominated by biomass burning (Ruellan et al., 1999;Formenti et al., 2003).Using aerosol mass concentrations together with measurements of aerosol optical properties (Magi et al., 2008), mass scattering cross sections (MSC) and mass absorption cross sections (MAC) for the aerosol species were derived.The measurement-based MSC and MAC for carbonaceous aerosol species (Table 6) all agreed with theoretical values of MSC and MAC predicted by Mie theory and with literature-based values.
The results presented in this study support the theory that the organic carbon (OC) component of an aerosol dominated by biomass burning particles absorbs a significant fraction of visible radiation.Climate models rely on a variety of assumptions about the basic properties (size, composition, refractive index, mixing state) of black carbon (BC) that are inconsistent with recent research (Bond and Bergstrom, 2006), and recent results (e.g.Kirchstetter et al., 2004) suggest the same is true for OC.As shown in this study, carbonaceous aerosol mass accounts for 54-83% of the aerosol mass over southern Africa, so assumptions about BC and OC are particularly important over biomass burning regions.The results presented here support the findings of Bond and Bergstrom (2006) regarding BC and those by Kirchstetter et al. (2004) for OC.Wild (2008) points out that climate models contributing to the IPCC 4th Assessment Report do not adequately account for absorbing aerosol, and most (see Meehl et al., 2007) do not even include OC and BC as forcing agents in climate simulations.Of the IPCC models that do include OC and BC, only the NASA GISS model (Liu et al., 2006) uses a treatment of OC absorption that is consistent with results by Kirchstetter et al. (2004).Both the NOAA GFDL (Magi et al., 2009) and the MPI ECHAM5-HAM (Stier et al., 2005) models use imaginary refractive indices for OC that are about five times less than suggested by current measurement-based results (Kirchstetter et al., 2004).The implications of more strongly absorbing OC on global and regional climate needs to be investigated, but as Wild (2008) showed, it is clear that the surface radiation budget cannot be properly simulated until we better understand basic aerosol properties and assimilate this understanding into climate models.
Models need to examine both the assumed properties of carbonaceous aerosol species and the subsequent mass apportionment in biomass burning regions to fully understand any discrepancies (e.g.Liu et al., 2006;Magi et al., 2009) that may exist.Results in this study will improve the understanding of both of these issues.In future work, the wavelength dependence of the mass scattering and mass absorption cross sections over southern Africa for the major aerosol species simulated in climate models will be discussed.This will provide further constraint on assumptions required by models to simulate the biomass burning component of the global aerosol.

Figure 1 .
Figure 1.Locations of filter samples collected during the SAFARI-2000 field cam (black crosses) and used in this study.Colors are mean (2001-2007) August and Sept fire counts derived from Moderate Resolution Imaging Spectroradiometer (MO measurements (Collection 5) in units of fires per half degree gridcell.

Fig. 1 .
Fig. 1.Locations of filter samples collected during the SAFARI-2000 field campaign (black crosses) and used in this study.Colors are mean (2001-2007) August and September fire counts derived from Moderate Resolution Imaging Spectroradiometer (MODIS) measurements (Collection 5) in units of fires per half degree gridcell.

Figure 2 .
Figure 2. Apportionment of PM 2.5 aerosol mass concentrations for (a) extratropical southern Africa and (b) tropical southern Africa.Shown are the median percent contributions of organic matter (OM), black carbon (BC), ammonium nitrate (AN), and ammonium sulfate (AS). 33

Fig. 2 .
Fig. 2. Apportionment of PM 2.5 aerosol mass concentrations for (a) extratropical southern Africa and (b) tropical southern Africa.Shown are the median percent contributions of organic matter (OM), black carbon (BC), ammonium nitrate (AN), and ammonium sulfate (AS).

Fig. 3 .
Fig. 3. Correlation of mass concentrations (µg/m 3 ) of (a) organic carbon (OC), (b) black carbon (BC), (c) nitrate ions (NO 3 ), and (d)sulfate ions (SO 4 ) with particulate matter with diameters less than 2.5 µm (PM 2.5 ).Listed in the figures are the slopes and 95% confidence interval of the linear regression, the correlation coefficients (r 2 ), and the p-values (p) of the correlation.

Figure 4 .
Figure 4. Apportionment of (a.) scattering and (b.) absorption at a wavelength of 550 nm in extratropical southern Africa.Shown are the median percent contributions (with the interquartile range) of organic matter (OM), black carbon (BC), ammonium nitrate (AN), and ammonium sulfate (AS). 35

Fig. 4 .Figure 5 .
Fig. 4. Apportionment of (a) scattering and (b) absorption at a wavelength of 550 nm in extratropical southern Africa.Shown are the median percent contributions (with the interquartile range) of organic matter (OM), black carbon (BC), ammonium nitrate (AN), and ammonium sulfate (AS).

Fig. 5 .
Fig. 5. Apportionment of (a) scattering and (b) absorption at a wavelength of 550 nm in tropical southern Africa.Shown are the median percent contributions (with the interquartile range) of organic matter (OM), black carbon (BC), ammonium nitrate (AN), and ammonium sulfate (AS).

Table 1 .
Characteristic properties of the extratropical and tropical

Table 3 .
Results of simultaneous filter samples showing aerosol mass concentrations (µg/m 3 ) determined using one of the three flow meters on the research aircraft.Flow Meter 2 had a reasonable calibration, while Flow Meters 1 and 3 were miscalibrated.

Table 5 .
Mass concentrations (µg/m 3 ) of PM 2.5 and species contributing to PM 2.5 (OC, BC, SO 4 , NO 3 , and SVOC) samples representative of the extratropics and the tropics of Southern Hemisphere Africa.Statistics are presented as median and interquartile range (IQR).OM is calculated as 1.4*OC in the extratropics, and 2.1*OC in the tropics (see discussion in text).

Table 6 .
The mass scattering cross sections for organic carbon (MSC OC ), black carbon (MSC BC ), sulfate (MSC SO 4 ), and nitrate (MSC NO 3 ) and the mass absorption cross sections for organic carbon (MAC OC ) and black carbon (MAC BC ), as derived from measurements and from Mie theory.The units are m 2 /g and MSC and MAC are at a wavelength of 550 nm.

Table 7 .
Aerosol characteristics used to calculate Mie mass scattering cross section and mass absorption cross section.The geometric mean diameter (D g ) and geometric standard deviation (σ g ) describe the lognormal aerosol size distribution.The real and imaginary refractive indices (m real and m imag ) are at a wavelength of 550 nm.