Red-ox speciation and mixing state of iron in individual African dust particles
Abstract
[1] The Fe distribution in African dust particles collected in Senegal (North-Western Africa) during the African Monsoon Multidisciplinary Analysis Special Observation Period 0 (AMMA-“SOP 0,” February 2006) was assessed using individual particle analysis (Scanning and Transmission Electron Microscopy respectively equipped with X-ray Spectrometry (SEM-EDX) and Electron Energy Loss Spectrometry (TEM-EELS)). Senegal is not a dust source area; the chemical composition of collected dusts indicates that they originate primarily in the North-Western Sahara, which is consistent with previous studies of the area. Fe can be present inside dust particles as a substitution element in the crystalline lattice of aluminosilicate, but a high proportion (62%) of aluminosilicate Fe-containing particles are also found as an internal mixture of aluminosilicate with Fe oxide grains (including both oxide and hydroxide species). The 3D structure of such particles obtained by tomography reveals that these Fe-rich inclusions are often found at the surface of aluminosilicate particles but that some are also included inside particles. These Fe oxide grains can result from crustal earth or atmospheric processes during long-range transport. FeIII is dominant in both the aluminosilicate matrix and the Fe oxide grains (FeIII/Σ Fe ratio = 76.8% and 90.0%, respectively, on average), with notable heterogeneities of Fe valence inside grains at a nanometer scale.
Key Points
- Aluminosilicate particles are mixed with small grains (<100nm) of Fe oxides
- Fe grains are found at the surface and in aluminosilcate particles
- Fe is essentially Fe(III), but Fe(II) is also found in the Fe grains
1. Introduction
[2] Iron (Fe) is a major chemical element in atmospheric aerosols and influences the optical properties of mineral dust [Derimian et al., 2008]. In addition, mineral dust acts as a vector for micronutrients, such as Fe, a key element in marine biogeochemistry [Boyd et al., 2010; Jickells et al., 2005; Johnson et al., 2003]. Long-range transport of mineral dust that primarily originates from large desert areas in the world is the main source of Fe for surface waters [Mahowald et al., 2005]. In this context, West Africa is a strategic zone to characterize Fe-containing dust advected over oceanic regions, insofar as Saharan dust plumes directly influence atmospheric Fe input to the North Atlantic Ocean [Fordham and Norrish, 1979; Sarthou et al., 2003]. As part of the AMMA (African Monsoon Multidisciplinary Analysis) international research project, an intensive field campaign called “Special Observing Period” (“SOP 0”) was carried out in West Africa during the dry season (February 2006), near M'Bour, Senegal. The aim of the ground-based sampling experiment was to determine the chemical composition of dust, with particular emphasis on the mixing state of aerosols. Collected mineral dust, carbonaceous and marine particles were predominantly found externally mixed, i.e., not present together in the same particles. However, binary internally mixed particles (dust/carbonaceous, carbonaceous/marine and dust/marine mixtures) accounted for a significant fraction (between 10.5% and 46.5%) of analyzed particles [Deboudt et al., 2010]. In the present paper, we investigated the variable composition of individual dust particles, with a special focus on the chemical form of Fe. Indeed, many recent works have demonstrated that Fe-bearing aerosols may have variable aqueous solubilities related to Fe chemical speciation [Baker and Croot, 2010; Baker et al., 2006; Mahowald et al., 2009; Okin et al., 2011; Paris et al., 2010; Schroth et al., 2009]. More particularly, the solubility of Fe-bearing particles significantly increases with the ferrous (FeII) content of those particles [Cwiertny et al., 2008; Trapp et al., 2010; Upadhyay et al., 2011]. The chemical composition of Fe-containing particles is studied here using transmission electron microscopy (TEM) coupled with electron energy loss spectroscopy (EELS). While the TEM-EELS technique has a lower energy resolution than other techniques recently used to investigate the red-ox speciation of Fe in aerosols, such as synchrotron-based X-Ray spectroscopies [Majestic et al., 2007; Oakes et al., 2012; Takahama et al., 2008; Takahashi et al., 2011], the implementation of TEM-EELS analyses is quick and inexpensive compared to synchrotron techniques and provides relevant results in terms of Fe speciation, with the advantage of a high spatial resolution.
2. Materials and Methods
2.1. Aerosols Sampling
[3] Sampling experiments were performed during the “dry-season” (SOP-0), from February 2 to February 15, 2006, South of M'Bour (14°23′38″N, 16°57′32″W), Senegal, at a ground-based station of the AMMA network. M'Bour is an African coastal city of 180,000 inhabitants, located 90 km south of Dakar. Atmospheric particles were collected by a cascade impactor (Dekati™ PM-10) containing two consecutive stages to separate coarse and fine fractions with equivalent cut-off diameters of 2.5 and 1μm, each with 50% of collection efficiency. Although the smallest cut-off diameter was 1μm, particles with diameters down to 300 nm were collected and analyzed [Deboudt et al., 2010]. Although the collection efficiency of submicrometer particles is low, many particles were collected due to their high number concentration in the air. Sampling was performed each day from approximately 12H00 to 13H00 UT, before the eventual formation of a sea breeze in the afternoon, which could increase the potential impact of local emissions. Particles were impacted on a boron substrate specifically developed for the purpose of particle microanalysis [Choël et al., 2005]. Transmission Electron Microscopy (TEM) grids with a Lacey carbon film were fixed on boron substrates for TEM-EELS analysis.
[4] Taking into account PM10 and Black Carbon (BC) concentrations and meteorological data acquired at M'Bour during the entire sampling campaign [Deboudt et al., 2010], six samples from the “SOP 0” period were selected, avoiding samples influenced by local emissions from the densely populated West African coast. These six samples (February 2, 3, 4, 13, 14 and 15, 2006) correspond to the highest PM10 concentrations during the “SOP 0” period (on average 107.5 μg.m−3) and are considered representative of the atmospheric matter leaving the coastal Sahelian zone and transported over the Atlantic Ocean during this period. An additional sampling experiment that corresponded to a huge dust event (mass concentration >1500 μg.m−3) was performed with the same methodology at M'Bour on May 22, 2009.
2.2. Automated SEM-EDX Analysis
[5] Single-particle analysis was performed with a LEO™ 438VP SEM, equipped with an ultrathin-window energy-dispersive X-ray detector. The microscope chamber was equipped with an oxygen radical source device (Evactron® RF plasma cleaner, XEI Scientific) to avoid organic carbon contamination from the pumping system and from samples. Automated particle analysis was run using the commercially available Link ISIS Series 300 Microanalysis System (Oxford Instruments®). Electron images were acquired using a backscattered electron detector. To limit the impact of particle size on the accuracy of morphological parameters, images were acquired with a magnification of ×4000 and ×6000 for impaction stages with equivalent cut-off diameters of 2.5 and 1μm, respectively. The pixel sizes were 90 nm and 60 nm, respectively. Electron images were converted to binary images using a preset gray level threshold to outline particles against the substrate. Each particle, considered a sphere, was morphologically characterized by its equivalent diameter calculated from the area of its two-dimensional projection. Only particles with diameters greater than 200 nm were considered. An X-ray spectrum of each particle was then automatically collected by scanning the electron beam over its projected area. X-ray spectra were collected with an acquisition time of 20 s at an accelerating voltage of 15 kV and a probe current adjusted to 200 pA with a picoamperometer (Keithley Instruments®). Net X-ray intensities of elements (Z > 4) were obtained by nonlinear least squares fitting of spectra [Vekemans et al., 1994]. Elemental intensities were then converted into atomic concentrations using a reverse Monte Carlo quantitative program [Ro et al., 2003]. Average errors obtained with this quantification method were better than 2.5 wt%, with effective correction of geometry effects over the entire size range (0.2–10 μm). This analytical methodology was successfully tested on airborne particulate matter that consisted of a complex heterogeneous mixture of particles [Choël et al., 2007]. To obtain a good precision of the relative abundance values for minor particle groups with relative abundance lower than 5%, a minimum number of 1,000 particles per impaction stage were analyzed in each sample, as was optimized in a previous study [Choël et al., 2010].
[6] The identification of individual particles was first based on their elemental composition obtained by SEM-EDX data, according to a procedure developed inDeboudt et al. [2010]. Marine, carbonaceous and metallic particles (Cr, Mn, Ni, Zn or Pb-rich) were removed from the data set because terrigenic mineral particles are only considered as aluminosilicate (including Mg aluminosilicates and silica), calcite, gypsum, Fe oxides, quartz, titanium oxide, mix aluminosilicate/Ca, mix aluminosilicate/Fe oxide and other terrigenic mixture particles. Excluding internally mixed particles, the mineral particles collected at M'Bour during the “SOP 0” account for, on average, more than 78% of externally mixed particles [Deboudt et al., 2010]. The category identified as “other terrigenic mixtures” includes particles composed of binary or ternary mixtures containing aluminosilicate, calcite, gypsum, Fe oxide, quartz and titanium oxide compounds, apart from Ca-enriched and Fe-enriched aluminosilicate particles.
2.3. TEM-EELS Analysis
[7] The structure, morphology and chemical composition of the most abundant types of particles containing Fe were manually analyzed by transmission electron microscopy (TEM) and electron energy loss spectroscopy (EELS). When transmitted through the sample, the electrons of the TEM beam may suffer an energy loss, i.e., an inelastic scattering. This loss may arise from excitation of a core energy level to the unoccupied density of state above the Fermi level and can be used to investigate the local chemistry [Egerton, 2009; Egerton and Malac, 2005]. The EELS Fe L2,3 edges correspond to excitations from the 2p63dn Fe ground state to the 2p53dn+1 states, where n = 5 for FeIII and n = 6 for FeII. The Fe L2,3edges of a range of minerals showed a constancy of edge shapes for a given oxidation state because the crystallinity of the sample has a low effect on these edge shapes dominated by quasi-atomic transitions with only minor modifications from the solid state environment [Calvert et al., 2005]. Thus, comparing the Fe L2,3 edges of an unknown sample with those of pure FeII and FeIII references (natural siderite and hematite, respectively) allows for determining the FeIII/ΣFe ratios in sample using Garvie and Buseck's [1998] method. The three spectra were scaled to each other by normalizing the edges to the area under the continuum well above the L2,3 white lines so that the spectral intensities reflect the ratio of FeII and FeIIId-holes. The Fe L2,3 edges were then obtained by subtraction of the continuum using a simple straight line function from 705 to 730 eV. Fe L2,3 edges of the unknown sample were then reconstructed by a linear combination of the two references, leading to the FeIII/ΣFe ratio in the sample. The quality of the fitting procedure was evaluated by the size of the residual. This method is robust and showed an error range of ±10% [Calvert et al., 2005]. The images and EELS spectra were obtained at 200 kV using a JEOL-2010F TEM equipped with a Gatan GIF system and higher spatial resolutions were obtained with a STEM VG equipped with a liquid nitrogen cooled sample holder and a home-developed EELS spectrometer. EELS data were routinely processed using standard GATAN software and then exported to the IGOR fitting package for further analysis. For elemental analysis, EELS spectra were acquired using a 0.5 eV/ch dispersion leading to an energy range of approximately 510 eV per spectrum. The energy loss functions were then scanned from 230 to 1950 eV to detect as many edges as possible. For Fe speciation analysis, EELS spectra were acquired using a 0.1 eV/ch dispersion leading to an energy range of approximately 102 eV per spectrum.
[8] Three-dimensional reconstruction of the local shapes, orientations and positions of regions of interest inside atmospheric particles were obtained by combining EELS Energy Filtered imaging with tomography. For this process, a tilt series of EF-TEM images were recorded, ranging from −55 to +46 degrees in 1 degree steps according to the Saxton method [Saxton et al., 1984] and using a FEI Tecnai (G2–20 twin, LaB6 filament, 200 kV) microscope equipped with a Gatan Energy Filter (GIF 2000). Digital processing of the images and visualization of the resulting tomographic reconstructions were carried out using Gatan 3D Visualization Software.
3. Results and Discussion
3.1. Overall Composition and Main Origin of Dust in West Africa
[9] The individual particle analyses of mineral dust collected at M'Bour during the “SOP 0” were extracted from the data set and the average chemical composition of these dust particles is presented in Figure 1. As expected, aluminosilicate, calcite and quartz constitute major compounds, but Ca-enriched and Fe-enriched aluminosilicate particles were also observed (aluminosilicate/Ca and aluminosilicate/Fe oxide mixtures, respectively) with a significant relative abundance (i.e., low percentages for particles >5μm to approximately 23% for the smallest particles).
[10] The mineral composition of airborne dust depends on the soil type in the source region. African dust is primarily composed of minerals such as quartz, feldspars, clays (mainly illite and kaolinite) and calcite [Formenti et al., 2011; Glaccum and Prospero, 1980], with a great variability in their relative abundance depending on the considered source area. Mineral dust transported from deserts of the North Sahara is rich in calcite and dolomite, whereas mineral dust emitted from the Central Sahara is rich in Fe oxides [Formenti et al., 2003]. In our study, collected fine dust originating from the North-Western Sahara is indicated by the presence of a significant fraction of calcite particles (approximately 9.3% of submicrometer particles) (Figure 1)and a very low fraction of pure Fe oxide particles (<1%; Figure 1). This result is also in agreement with the relative abundances of hematite and calcite particles collected by Kandler et al. [2009]during dust storms in Morocco (North-Western Sahara).
[11] Clays can also be used as source tracers: for example, illite is a ubiquitous mineral, whereas the distribution of kaolinite, like many clay minerals, is a function of large-scale weathering leading to a latitudinal distribution, such as in deep-sea sediments [Chamley, 1989]. Consequently, the illite-to-kaolinite mass ratio progressively decreases from 2.3 to 0.1 from the North-Western Sahara to central Sahel and from west to east [Caquineau et al., 1998, 2002; Molinaroli, 1996; Reid et al., 2003]. Mineralogical compositions cannot be directly determined by SEM-EDX analysis, but considering the chemical composition of minerals (Al4Si4O10(OH)8 and K<2Al4[Si>6Al<2]O20(OH)4.nH2O for kaolinite and illite, respectively [Brindley and Brown, 1980], they can be roughly distinguished by their elemental ratios if they do not mix internally and do not mix with other species (i.e., metal oxides). Ideally, the Al:Si ratio is 1 for kaolinite and approximately 0.5 for illite. However, the latter figure is highly variable, due to possible substitutions of Al by Fe or Mg in octahedral sites [Deer et al., 1992]. In addition, considering the possible introduction of K within the interlayer space of phyllosilicates, illite can be distinguished from kaolinite by significant fractions of K, Fe and Mg. As a result, the relative abundance of aluminosilicate particles for Al/Si and (K+Mg+Fe)/(Al+Si) atomic ratios (Figure 2) provides information on the relative proportion of illite to kaolinite in individual particles and on the possible origin of clay materials according to individual particle size. All of our analyzed particles identified as aluminosilicate particles are reported in Figure 2. Two main findings are observed: (1) Al/Si ≈ 0.9–1 and limited concentrations of (K+Mg+Fe) correspond to the kaolinite compositions, and (2) Al/Si ≈ 0.7 and a higher fraction of K+Mg+Fe can be attributed to illite particles. For the fine fraction (Figure 2a), the aluminosilicate are mainly composed of kaolinite particles. Some particles have the ideal Al/Si ratio of 1 and a very low K+Mg+Fe/Al+Si ratio and may correspond to an ideal kaolinite composition. Al/Si ≈ 0.9 corresponds to kaolinite particles with Al deficiency due to higher K+Mg+Fe substitution. A small contribution of kaolinite with an ideal Al/Si ratio of 1 and some K+Mg+Fe contribution may correspond to perfect kaolinite superimposed with some Fe oxides. For the coarse fraction (Figure 2d), the majority of particles are now illite, and the remaining fraction of kaolinite shows systematic deviation from the ideal Si/Al ratio. In conclusion, although illite is not the only clay for dusts collected at M'Bour, the submicronic fraction is dominated by kaolinite particles, while supermicrometer particles are primarily illite particles (notably for larger particles with equivalent diameters higher than 2.5 μm). Considering the relative size fractions found on the impactor stages, the mass of illite particles is higher than kaolinite particles and the illite to kaolinite mean mass ratio is certainly >1. Compared to data found in the literature [Caquineau et al., 1998; Caquineau et al., 2002; Molinaroli, 1996; Reid et al., 2003], this study strongly confirmed that collected dust originates from the North-Western Sahara, which is consistent with the main geographic origin of air masses ending at M'Bour (Figure 3).
[12] PM-10 mass concentrations measured during “SOP 0” at M'Bour were moderate (from 10 to 500μg.m−3) [Flament et al., 2011], meaning that the “SOP 0” sampling campaign was not carried out under the influence of Saharan dust storm events, which constitute the major episodes of dust export to the North Atlantic Ocean. In the case of dust storms, are particle types the same as those found during “SOP 0”? To answer this question, we performed an additional sampling during a huge dust event (PM-10 mass concentration >1500μg.m−3) in May 2009 using the same methodology. The back-trajectory of air masses ending at M'Bour (Figure 3) indicates that for this additional experiment, collected aerosols come from the North-Western Sahara, as does dust collected during “SOP 0.” Analyzed particles have similar characteristics: aluminosilicate particles are predominant, and the relative abundance of calcite particles is significant (6.1%), while the relative abundance of pure Fe oxide particles is very low (1.2%). Thus, although based on the assessment of a single event, we believe that dust collected at M'Bour during the “SOP 0” is representative of dust exported to the North Atlantic Ocean during Saharan dust storm events in winter/spring. This conclusion is confirmed by other works [Chiapello et al., 1997; Hoornaert et al., 2003; Kandler et al., 2007; Skonieczny et al., 2011] in which mineral dust transported on the West African margin during prominent winter/spring events is associated with various geographic sources from a large area covering Mauritania, Mali and South Algeria.
3.2. Fe-Rich Aluminosilicate Particles
[13] Regardless of particle size class, Fe is present in mineral dust as internally mixed with aluminosilicate rather than as pure Fe oxide particles (Figure 1). The identification of atmospheric particles was based on their elemental composition obtained by SEM-EDX according to the criteria presented inFigure 4. Among all of the analyzed atmospheric particles (12,672 particles), Fe-containing particles are classified into 3 categories: aluminosilicate, Fe oxides and mix aluminosilicate/Fe oxide particles. At times, Fe can be included in binary or ternary internally mixtures: these particles are here classified as “other particles” (except for aluminosilicates/Fe oxide mixtures, the most abundant internally mixed particles containing Fe). The relative abundance of Fe-containing particles is higher in submicrometer particles, which are easily transported over great distances and, consequently, greatly influence ocean geochemistry. Mineralogy of African dusts and soil surfaces in source regions is well documented [Avila et al., 1997; Caquineau et al., 2002; Claquin et al., 1999; Glaccum and Prospero, 1980; Molinaroli, 1996; Moreno et al., 2006]. The common minerals containing Fe in a significant fraction are listed in Table 1. Considering the Fe content in these minerals and the accuracy of automated SEM-EDX analysis [Choël et al., 2005], aluminosilicate particles with an Fe content higher than 6 at.% cannot be considered as pure aluminosilicate particles but as mix aluminosilicate/Fe oxide particles. They are composed of free Fe oxides in fine-grained materials associated with aluminosilicate particles and cannot be substitution Fe in the crystalline matrix of aluminosilicate (Figure 4). This type of Fe oxide grain has already been reported in clays and African soils [Fordham and Norrish, 1979; Greenland et al., 1968; Sei et al., 2004, 2006], but has rarely been studied in fine atmospheric dust particles [Lieke et al., 2011; Scheuvens et al., 2011]. However, aluminosilicate particles with an Fe-content lower than 6 at.% cannot be systematically considered as substitution Fe in the crystalline matrix of aluminosilicate, as it could also be an Fe oxide inclusion. Further experiments concerning the Fe distribution in these particles will be developed in the next section.
| Minerals/Mineral Groups | Type | Formula | Fe Contenta |
|---|---|---|---|
| illites | (K,H3O)(Al, Mg, Fe)2(Si,Al)4O10[(H2O),(OH)2] | <2 at. % | |
| montmorillonites (smectite) | Clay minerals | (Na,Ca)0,3(Al,Mg)2Si4O10(OH)2,nH2O | <1 at. % |
| palygorskite | (Mg,Al)2Si4O10(OH).4H2O | <2 at. % | |
| feldspar | Silicate | (Na, Ca, K)Al(Al,Si)3O8 | <3 at. % |
| hematite | Fe2O3 | 40 at. % | |
| goethite | Oxides | FeO(OH) | 33 at. % |
| magnetite | Fe3O4 | 43 at. % |
- a Fe contents are calculated for oxides without considering H, not detected by SEM-EDS or fromDeer et al. [1992] for silicate and clay minerals.
[14] On average, pure Fe oxides particles are scarce and Fe-rich particles are essentially mix aluminosilicate/Fe oxide particles (Figure 1), notable for the fine fraction of dust (diameter <1 μm). The shape factor (Sf) associated with these particles seems to be influenced by particle size (Figure 5), which provides information about the particle roughness and discontinuities. That is Sf = 1 for a spherical particle and is >1 for irregular particles with rough edges and aggregates. It is linked to the 3D-aspect ratio of particles which has a direct impact on their optical properties and can induce high differences in radiation fluxes [Torge et al., 2011]. In our case, the mean Sf is lower for fine than for coarse particles, indicating that they are more regular in shape. This finding is consistent with others' observations of African dust [Chou et al., 2008; Kandler et al., 2011; Koren et al., 2001]: irregular aggregates are encountered more often in the coarse fraction than in the fine fraction. Moreover, although the number of analyzed Fe oxide particles is relatively low (76 particles), their Sf seems to be lower than the Sf obtained for Fe-rich aluminosilicates, indicating that the Fe oxide particles are more regular in shape than are aluminosilicates.
[15] Although particle types observed during the May 2009 dust storm event are the same as those collected during “SOP 0,” the dust size distribution is extremely different. Only 95 analyzed particles identified as dust (6.1%) have a diameter <1 μm, which is low compared to “SOP 0” (Figure 1), but consistent with optical data registered at M'Bour during storm events. Huge dust events are systematically associated with lower Angstrom exponents (870 nm/440 nm) (i.e., large particles) than for moderate dust events [Derimian et al., 2008]. Considering all size fractions, Fe oxides and mix aluminosilicate/Fe oxide particles each account for less than 2% of dust during the May 2009 storm event. Dust is essentially composed of aluminosilicate, quartz and calcite particles (76.7%, 9.2% and 6.1% of analyzed particles, respectively), which is relatively similar to the “SOP 0” coarse fraction (Figure 1). During “SOP 0,” the mean diameter of Fe oxides and mix aluminosilicate/Fe oxide particles (0.77μm and 0.85 μm, respectively) is lower than the mean diameter of aluminosilicate particles (1.43 μm).
3.3. Distribution of Fe in Aluminosilicate Particles
[16] Insofar as aluminosilicate particles are the most abundant type of Fe-containing particles in the case of a huge dust event (see above), it is essential for marine biogeochemistry to assess the Fe distribution and speciation in these particles. Among the 43 typical dust particles randomly selected and manually analyzed by TEM-EELS, only 5 particles did not contain Fe (Figure 6). Fe is detected in approximately 90% of the aluminosilicate particles, which accounts for a higher fraction than the relative abundance of mix aluminosilicate/Fe oxide particles observed by automated SEM-EDX (Figure 1). There is no quantitative determination of Fe content inside aluminosilicate particles by EELS, but all of these observations indicate that a high fraction of particles identified as pure aluminosilicate particles by automated SEM-EDX likely contains nano-inclusions of Fe oxides and should be identified as mix aluminosilicate/Fe oxide particles, even if the Fe content is low (less than 6 at.%).
[17] The Fe distribution in the Fe-containing aluminosilicate particles has been demonstrated by TEM-EELS by comparing of spectra acquired at different areas of the observed particles.Figure 7illustrates the different types of Fe-containing particles observed by TEM-EELS. The O-K edge of the pure Fe-oxide particle (Figure 7a) shows an initial low intensity peak at ≈525 eV energy loss, which is not observed for Fe-free aluminosilicate particles or for Fe-containing aluminosilicate particles with a homogeneous distribution of Fe (Figures 7b and 7c). This pre-peak attributed to hybridization of oxygen with neighboring Fe atoms is generally not found for silicates characterized by more ionic bonds than in Fe oxides and increases with FeIII content in these oxides [Calvert et al., 2005; Laribi et al., 2007]. The O-K edge of aluminosilicate particles with a heterogeneous Fe distribution (Figure 7d) generally shows a low intensity pre-peak at ≈525 eV energy loss, similar to that observed for Fe oxide particles (Figure 7a), indicating the presence of pure Fe oxide zones (including both oxide and hydroxide species) inside the particle. All of our observations reveal that Fe distribution is heterogeneous in the majority of Fe-containing aluminosilicate particles (62%). Moreover, most of atmospheric aluminosilicate particles are normally mixed with grains of Fe oxide materials. The 3D structure of such grains has been established by Electron Energy-Loss tomography. Although the inclusion of Fe oxides is often found at the surface of aluminosilicate particles, some Fe oxides are found inside particles, as illustrated inFigure 8. Fe inclusions 1 and 3 (in red) are only partially visible, depending on the observation angle; they are localized inside the aluminosilicate particle when #2 is completely visible in Figure 8fso is localized at the surface of the particle. The size of these Fe inclusions is relatively small, typically approximately 100 nm. The localization of these Fe inclusions inside or at the surface of particles has been confirmed by 3D reconstruction of particles from a tilt series of EF-TEM images (not presented here). Despite the interest this localization has for future optical modeling, the quantification of the relative fraction of particles with Fe at the surface versus Fe inside the particle were not performed due to a poor representation of a fraction calculated from the few 3D reconstructions performed for this work.
[18] This type of heterogeneous Fe-aluminosilicate particle has already been directly observed in tropospheric aerosols [Jeong, 2008; Kandler et al., 2007; Oakes et al., 2012]. Shi et al. [2009]reported the formation of ferrihydrite micro-aggregates through simulation experiment of cloud processes with African dust. However, the Fe grains in mineral dust observed in our study are likely not the result of cloud processes because these inclusions are observed for samples collected on February 2, 3 and 4, 2006, for which the collected air masses were associated with relative humidity (RH) not exceeding 35% during the last 120 h before sampling (data from HYSPLIT model, http://www.arl.noaa.gov). At this low RH, cloud processes cannot be considered. Here, the Fe nano-grains in mineral dust are likely the result of oxidation of Fe-aluminosilicate in the crustal earth as a pedogenic process like rubification [Ben-Dor et al., 2006; Cornell and Schwertmann, 2003; Deer et al., 1992]. These processed Fe-aluminosilicate particles characterized by the presence of Fe oxides are typical of old, highly weathered and oxidized soil clay particles [Fordham and Norrish, 1979; Greenland et al., 1968].
3.4. Variability of Fe Red-Ox Speciation in Fe-Rich Aluminosilicate Particles
[19] More than 60 EELS spectra of the Fe L2,3edge were acquired with a high energy resolution to estimate the Fe valence in different areas of typical heterogeneous Fe-rich aluminosilicate particles. The general conclusion of these experiments is that Fe speciation is highly variable in these particles, although FeIII is dominant: the average FeIII/ΣFe ratio is 76.8 ± 14.7% (1 SD). This variability is observed for both the aluminosilicate matrix and the Fe oxide grains as illustrated in Figure 9. AlSi A and AlSi Bare two areas corresponding to an Fe-rich aluminosilicate crystalline lattice, which is confirmed by the absence of pre-peak at approximately 525 eV energy loss of the O-K edge. Although these two areas are distant from less than 50 nm, their associated Fe speciation is clearly distinct (i.e., 72.2% and 87.2% of FeIII, respectively). Therefore, these spectra suggest that both FeII and FeIII were substituted into aluminosilicates as expected. The state of Fe at the micrometer scale in clays is highly variable, strongly dependent on the disorder of the crystalline matrix [Laribi et al., 2007] and redox biotic or abiotic processes [Anastácio et al., 2008; Cornell and Schwertmann, 2003].
[20] Concerning the Fe oxide grains, although they are essentially composed of FeIII (approximately 90% in Figure 9), heterogeneities of Fe valence were also observed inside them, when Spectrum Images (SPIM) were acquired at high magnification. In Figure 10, acquisitions of the Fe L2,3 edge along a line crossing the Fe oxide grain clearly indicate a predominance of FeIII at side A of the grain (FeIII/ΣFe = 67.0 ± 3.1%) and FeII at side B (FeIII/ΣFe = 32.9 ± 2.9%). Fe is not detected in the neighboring aluminosilicate matrix. Unfortunately, it was not possible to determine the structural composition of these Fe inclusions in our analytical configuration, and it is unclear if they exist as amorphous or crystalline compounds. The presence of FeII is observed for particles collected in air masses associated for both dry (<35%) and humid (>90%) situations during the last 120 h before sampling and no clear tendency can be determined for the relative abundance of FeII according to relative humidity.
[21] In African desert source areas, Fe oxides are generally goethite (FeIII) or, to a lesser extent, hematite (FeIII) [Lafon et al., 2006] and the fraction of FeII is relatively low [Cwiertny et al., 2008; Schroth et al., 2009]. Consequently, the presence of FeIIin our sample could be due to atmospheric processes during long-range transport as photo-reduction [Zhu et al., 1997] rather than soil processes, even if this last type of processes cannot be completely excluded.
[22] Takahama et al. [2008] observed the reduced form of iron (FeII) on particle surfaces of Fe-containing atmospheric aerosols (16 of 63 atmospheric particles mapped by near-edge X-Ray absorption fine structure spectroscopy), suggesting atmospheric surface processing.Majestic et al. [2007] observed a slight decrease in the FeII/FeIIIratio in urban aerosol samples when they were artificially aged by illumination to represent long-range transport with low relative humidity conditions.Moffet et al. [2012] observed FeII-rich grains inside anthropogenic particles and suggested Fe photo-reduction by sulphate species. Similar processes could be responsible of the presence of FeII in Fe oxide grains of our collected aluminosilicate particles. Sulphur species were not observed in aluminosilicate particles, but anthropogenic carbonaceous species that could act as reductive agent were detected [Deboudt et al., 2010].
4. Summary
[23] Although mineral dust present in the African troposphere is essentially composed of externally mixed compounds (e.g., illite, kaolinite, calcium carbonate, gypsum, Fe oxides, silica, titanium oxide…), all of our observations reveal that the majority of atmospheric aluminosilicate particles are composed of internally mixed aluminosilicate and Fe oxide compounds. Fe can be present as substitution Fe in the crystalline matrix of aluminosilicate, but the existence of Fe oxide nano-inclusions has also been highlighted at the particle scale, even at low Fe-content levels(less than 6 at.%). The Fe distribution in the aluminosilicate Fe-containing particles was determined using TEM-EELS. The Fe distribution is heterogeneous in the majority of Fe-containing aluminosilicate particles, with the presence of pure Fe-oxide zones (including both oxide and hydroxide species) inside the particles, which is likely the result of oxidation of Fe-aluminosilicate particles in the crustal earth as pedogenic processes. Since the solubility of Fe oxides is strongly size dependant [Baker and Jickells, 2006; Cornell and Schwertmann, 2003], the small size of these Fe grains (some dozens of nanometer) in dust suggests that they are more likely to be solubilized after atmospheric processing and become bioavailable.
[24] The grains of Fe oxides are often found at the surface of aluminosilicate particles, but some are included inside particles. Fe speciation is highly variable in these particles, although FeIII is dominant (FeIII/ΣFe ratio = 76.8 ± 14.7% on average). This variability is observed for both the aluminosilicate matrix and Fe oxide nano-inclusions. The obtained EELS spectra suggest that both FeII and FeIII were substituted in aluminosilicates. Concerning the Fe oxide inclusions, although they are essentially composed of FeIII (approximately 90%), heterogeneities of Fe valence were also observed in them. Insofar as African desert sources contain goethite or hematite, the presence of FeIIin a relatively high abundance could be due to atmospheric photo-reduction during long-range transport rather than soil processes.
[25] As reported by Raiswell and Canfield [2012], the close association of clay minerals in natural systems with nanoparticles of iron (oxyhydr)oxides makes extremely difficult to disentangle these two sources of soluble Fe. These results should be considered in dissolution experiments, notably in the choice of standard minerals to represent minerals in natural dust sample.
Acknowledgments
[26] Based on a French initiative, the AMMA project was built by an international scientific group and is currently funded by many agencies in France, the U.K., the U.S. and Africa. It has been the beneficiary of a major financial contribution from the European Community's Sixth Framework Research Programme. Detailed information on scientific coordination and funding is available on the AMMA International website http://www.amma-international.org. We are especially grateful to the “Action Programmée Interorganismes” API-AMMA, the “Fédération de Recherche CNRS 1818” and the “Laboratoire d'Optique Atmosphérique – Université des Scienes et Technologies de Lille” for their special funding. The laboratory LPCA also participates in the Institut de Recherche en ENvironnement Industriel (IRENI) which is financed by the Communauté Urbaine de Dunkerque, the Région Nord Pas-de-Calais, the Ministère de l'Enseignement Supérieur et de la Recherche, the CNRS and European Fund for Regional Development (FEDER). Karine Deboudt expresses grateful thanks to Christian Colliex for his enthusiastic welcome at the LPS at Orsay in 2008–09. The authors thank Charlotte Skonieczny and Aloys Bory (Geosystemes, FRE CNRS 3298, Université Lille1, France) for providing the sample from the dust event in May 2009 and the management team of IRD-Senegal for facilities at the environmental research station of M'Bour. The TEM national Facility in Lille (France) is supported by the Conseil Regional du Nord-Pas de Calais, the European Regional Development Fund (ERDF), and the Institut National des Sciences de l'Univers (INSU, CNRS). Finally, the authors wish to thank the anonymous reviewers for their helpful comments.
