3.1. Mineral Dust and Biomass Burning Aerosols, and Their Mixing
 A summary of the mean elemental concentrations of the major constituents (Na, Mg, Al, Si, P, S, K, Ca, Ti, and Fe) observed at the ground and on the aircraft during the field campaigns is shown in Table 1. For sake of comparison with published data, elemental ratios with respect to Al are also shown. Mean values for ground-based and aircraft samples are similar. Higher mean concentrations and variability were obtained for the DODO2 campaign because of an extremely intense dust storm which was sampled on 23 August 2006.
Table 1. Mean Elemental Concentrations and Mean Elemental Ratios to Ala
| ||Concentrations (μg m−3)|
|Banizoumbou (Jan–Feb 2006)||AMMA SOP0/DABEX (Jan–Feb 2006)||DODO1 (Feb 2006)||DODO2 (Aug 2006)|
|Na||1 (1.0)||2 (1)||4 (4)||8 (14)|
|Mg||3 (1.5)||3 (2)||4 (5)||12 (25)|
|Al||18 (9)||17 (19)||9 (10)||66 (162)|
|Si||48 (25)||48 (48)||24 (29)||172 (415)|
|S||—||2 (1)||4 (5)||3 (6)|
|K||3 (2)||5 (4)||3 (4)||19 (46)|
|Ca||6 (4)||9 (9)||9 (19)||36 (82)|
|Ti||1.4 (0.6)||2 (2)||1 (1)||6 (15)|
|Fe||10 (5)||10 (11)||6 (8)||45 (109)|
| ||Ratios to Al|
|Banizoumbou (Jan–Feb 2006)||AMMA SOP0/DABEX (Jan–Feb 2006)||DODO1c (Feb 2006)||DODO2 (Aug 2006)|
|Na||0.05 (0.04)||0.06 (0.03)||0.5 (0.7)||0.3 (0.6)|
|Mg||0.14 (0.03)||0.2 (0.1)||0.3 (0.1)||0.2 (0.1)|
|Si||2.7 (0.3)||3.0 (0.6)||2.7 (0.2)||2.7 (0.4)|
|S||—||0.2 (0.3)||0.1 (0.1)||0.1 (0.3)|
|K||0.20 (0.05)||0.3 (0.2)||0.2 (0.1)||0.24 (0.06)|
|Ca||0.4 (0.1)||0.5 (0.3)||0.6 (0.6)||0.6 (0.2)|
|Ti||0.08 (0.01)||0.10 (0.04)||0.1 (0.1)||0.08 (0.01)|
|Fe||0.59 (0.06)||0.7 (0.3)||0.70 (0.08)||0.68 (0.04)|
 As demonstrated by Osborne et al.  and Johnson et al. , in wintertime the atmospheric column over western Africa was characterized by a multilayer aerosol structure. During winter dust was found close to the surface owing to transport at low level by the Harmattan winds, whereas persistent layers of biomass burning (BB) were found aloft, between 2 and 6 km. Conversely, dust was transported at few-km altitude in summertime [McConnell et al., 2008].
 On the basis of visual observations and online monitoring of the spectral behavior of aerosol scattering coefficient available on board the BAe-146, we were able to collect filter samples from layers representing different aerosol conditions.
 Dust was found ubiquitously in all of them, including in the elevated BB layers. Evidence for this is in the PIXE spectra for low-Z elements, where the signals of Al and Si were evident together with that of S and K, good tracers for aged BB [Gaudichet et al., 1995; Ruellan et al., 1999]. As a matter of fact, because of their large mass and despite their little number, crustal particles dominated the mass concentrations over biomass burning aerosols, which were numerous but small.
 Nevertheless, the relative contribution of mineral dust to the total aerosol load changed with aging of air masses. This effect is clear when looking at the linear correlation between elemental K and Al (Figure 1) for aircraft samples collected in dust layers, fresh BB layers (that is, within aircraft overpass of a smoke plume), and in the elevated BB layers. Only four samples were collected in fresh biomass burning. With this limitation in mind, the K-to-Al ratio was similar for fresh biomass burning and dust layers (0.16–0.18) and close to previous values obtained for mineral dust (∼0.2 [Chiapello, 1996; Formenti et al., 2003]) and to the crustal reference value (0.3) of Mason . The ratio was higher (K-to-Al = 0.52) in the elevated biomass burning layers, likely due to the depletion of the coarse mode mineral dust particles during transport and uplift in the atmosphere. As a consequence, the relative contribution of fine mode K-containing particles became more significant. Similar ratios have been found previously in smoke from smoldering biomass burning [e.g., Gaudichet et al., 1995; Ruellan et al., 1999].
Figure 1. Scatterplot of elemental concentrations of K and Al measured on board the BAe-146 aircraft during the AMMA SOP0/DABEX campaign. Data have been sorted by aerosol type: dust (circles), old BB (squares), and fresh BB (diamonds). The slope of the regression line and the square of the correlation coefficient (R2) are shown in parentheses.
Download figure to PowerPoint
 To illustrate the relative contribution of dust and BB to the aerosol load, we estimated the dust mass (DM) as sum of the major oxides (Na2O, MgO, Al2O3, SiO2, K2O, CaO, TiO2 and Fe2O3), and the BB aerosol mass (BBM) from measured total carbon (TC). Of course, this is a lower limit for BBM. An upper limit, taking into account the oxygen and hydrogen associated with organic matter as well as the inorganic fraction, was estimated to be 1.3 × TC by taking literature values of BC/TC, OC/TC and POM/OC [Ruellan et al., 1999].
 With these values, we found that the contribution of mineral dust to the aerosol mass load was overwhelming, independently on the nature of the observed air mass. The percent contribution of DM ranged from 93 (±7) % in dust layers to 72 (±16) % in the elevated BB layers. The percent contribution of DM in fresh BB was 91 (±9) %, as in dust layers. That is consistent with visual observations during flights, showing that several tenuous smoke plumes from localized small fires diffused almost vertically into a widespread regional haze, mainly composed of mineral dust. The regional character of this persistent and diffuse dust haze is testified also by the fact that debris of fossil diatoms were observed on almost all the filters collected in the surface BB layers, despite the fact that their major source, the Bodélé depression, was not always active during the measurement period. The contribution of mineral dust to the column aerosol load seemed therefore to be rather uniform in the boundary layer, and only decreased in the layers above.
 These observations are contrasting to those of Ruellan et al. , who found that, during the fire season, mineral dust accounted for only up to 37% of the aerosol mass in the boundary layer over Ivory Coast, around 4°N, that is, south of the Intertropical Convergence Zone (ITCZ), which is at approximately 8°N at the ground at this time of the year. All our observations have been performed above 8°N.
 Finally, the question can now be asked about mixing. The investigation by electron microscopy of several filter samples collected in mineral dust, fresh and aged biomass burning layers (not shown), are consistent in suggesting that dust and biomass burning aerosols were externally mixed. Chou et al.  have shown that biomass burning aerosols in dust layers are essentially K-S particles in the 0.1–0.6 μm diameter range, whereas mineral dust particles are seldom found below 0.4 μm diameter. These observations match those of Johnson et al.  based on analysis of particle number size spectra from optical counting.
 External mixing is consistent with the fact that Johnson et al.  and Osborne et al.  were able to deduce the optical properties of either mineral dust or biomass burning aerosols by linearly deconvoluting measurements performed in mixed layers.
 Although electron microscopy performed under vacuum is not suitable to investigate coatings of organic materials, it is worth mentioning that no evidence for organic coating of dust particles could be observed in any of the many samples investigated. On the contrary, transmission electron microscopy yields examples of sulfate coatings on soot particles (not shown). Coatings of mineral dust by organic carbon will be addressed more systematically in future work.
3.2. Elemental Composition of Mineral Dust, Including Regional Variability
 In this section we investigate the elemental composition of mineral dust. To make sure to avoid contamination, we only selected samples for which Al concentrations were higher than 20 μg m−3.
 Regarding ground-based samples, Rajot et al.  were able to isolate eighteen episodes of dust on the basis of the correlation between mass and number concentrations. Five of these episodes were of local origin; the remaining thirteen corresponded to dust advected from more distant sources. By coupling a statistical analysis (Principal Component Analysis (PCA)) of the temporal covariance of the elemental concentrations and calculated five-day back trajectories, Rajot et al.  were also able to show that the various episodes of advected dust had originated from different source regions.
 Thirteen filters were retained during episodes of advected dust and five during episodes of local production. For each of these samples, we calculated the mean composition of mineral dust as percent of the total mass estimated as sum of the measured concentrations for Na, Mg, Al, Si, K, Ca, Ti and Fe. These results are presented in Table 2.
Table 2. Mean Elemental Composition of Mineral Dusta
| ||Banizombou (Local)||Banizoumbou (Advected)|
|Mean (±std)||Variability (%)||Mean (±std)||Variability (%)|
|Na||0.9 (0.6)||65||1.1 (0.7)||62|
|Mg||2.1 (0.1)||7||3.0 (0.4)||15|
|Al||21.4 (0.6)||3||19 (2)||11|
|Si||53.5 (0.6)||1||54 (3)||4|
|K||3.6 (0.2)||7||3.5 (0.3)||10|
|Ca||4.6 (0.5)||10||7.5 (1.6)||21|
|Ti||2.3 (0.1)||3||1.4 (0.2)||15|
|Fe||12.3 (0.3)||3||10.9 (0.9)||9|
| ||AMMA SOP0/DABEX||DODO2 (B238)|
|Mean (±std)||Variability (%)||Mean (±std)||Variability (%)|
|Na||1.0 (0.4)||43||1.4 (0.1)||8|
|Mg||2.2 (0.2)||10||3.4 (0.4)||14|
|Al||20 (2)||8||19 (1)||5|
|Si||51 (2)||3||46 (1)||3|
|K||3.6 (0.2)||5||5.3 (0.1)||3|
|Ca||9.1 (0.5)||6||11 (2)||19|
|Ti||1.6 (0.2)||11||1.6 (0.1)||7|
|Fe||11.6 (0.6)||5||12.7 (0.6)||5|
 We first investigate ground-based data. The mean composition obtained for locally produced dust is 21.4% (±0.6) for Al, 53.4 (±0.6) for Si, 12.3% (±0.3) for Fe, 4.6% (±0.5) for Ca, 3.6% (±0.2) for K, 2.3% (±0.1) for Ti, 2.1% (±0.1) for Mg, and 0.9% (±0.1) for Na. Beside Na, which was above analytical detection limit only on two of the 5 samples, little variability (within 10%) was observed from sample to sample.
 The mean composition obtained for advected dust is 19% (±2) for Al, 54 (±3) for Si, 10.9% (±0.9) for Fe, 7.5% (±1.6) for Ca, 3.5% (±0.3) for K, 1.4% (±0.2) for Ti, 3.0% (±0.4) for Mg, and1.1% (±0.7) for Na. One can immediately note that the percent variability on each of the components is much larger (up to 21% for Ca).
 Furthermore, the composition of locally produced and advected dust is similar, yet different. Si is ubiquitous and found in identical proportions. On the contrary, locally produced dust has lower Ca and Mg contents, but higher Al, Fe and Ti than advected dust. A t-test confirms that these differences are not speculative but have statistical significance (95% significance level).
 The enrichment of local dust in Al, Fe and Ti is in agreement with the expected composition of Sahelian dust, dominated by calcium-poor clays, mainly kaolinite (Al4[Si4O10](OH)8) but also illite ((K0.7, Na0.2, Ca0.1)(Al3, Fe0.6, Mg0.4)(Si7, Al)O2(OH)4) [Caquineau et al., 1998], as well as iron-oxides such as hematite and goethite. The good correlation between Al, Fe and Ti (R2 = 0.79 and R2 = 0.76, for Fe versus Al and Ti versus Al, respectively) suggests that Fe and Ti are associated with clays. Fe is found in the crystalline lattice of clays, but also as iron oxides (such as hematite/goethite, Fe2O3/FeO-OH respectively), in the form of little spheres on the surface of clay particles [Lafon et al., 2006]. Ti oxides such as anatase (TiO2) are also found in mineral dust in association with clays. As a matter of fact, small individual particles of Ti and iron oxides were observed by electron microscopy on some of the aircraft filters [Chou et al., 2008]. Conversely, the advected dust is marked by enrichment in Ca and Mg. These elements were correlated (R2 = 0.67). Carbonates such as calcite (CaCO3) and dolomite (CaMg(CO3)2) are found in the North African deserts of Libya, Algeria, and Tunisia, thus calcium is a good tracers for Saharan dust [Lafon et al., 2006].
 We can now explore the variability in the mean composition of the advected dust which could be due to the different origin of the various episodes [Rajot et al., 2008]. To do so, we sorted the percent composition of each of the episodes accordingly to their Ca-content. Because Ca showed the largest percent variability (ranging from 5.3 to 10.1% of the estimated dust mass), this allows maximizing sample-to-sample differences. Each sample was associated at the corresponding five-day back trajectory. When doing so, samples having Ca content lower than 7% seemed to associate with air masses from within Niger, whereas samples for which the Ca percent value was larger than 7% rather associated with transport from Algeria, Libya and western Africa/Morocco. One sample (SOP0-4) is peculiar. This corresponded to the outbreak of mineral dust from the Bodélé depression that had also been detected by the lidar system installed in Banizoumbou [Heese and Wiegner, 2008]. The identification of the Bodélé as the source region is unequivocal owing to the numerous diatoms debris that appeared on the electron microscope images. Bodélé has been shown to be the major source region of such diatoms [Washington et al., 2006]. The composition of the sample attributed to the Bodélé source is significantly different (t-test, 95% significance level) from that of the other episodes, except that for Ca and Mg. In particular, the samples from Bodélé had the highest Si content (60%), signature of the fossil diatoms, and the lowest Al, K, and Fe content (15, 3.1 and 8.9%, respectively).
 This analysis was extended to the aircraft samples. Nine aircraft samples were retained for the AMMA SOP0/DABEX phase, which were collected mainly during three dedicated flights [Osborne et al., 2008; Chou et al., 2008], but also during straight and leveled runs (SLRs) of opportunity. Twelve samples were retained for DODO, 4 from DODO1 and the remaining 8 from DODO2. The DODO2 samples were mainly collected during two severe dust storms from western Sahara and Morocco [McConnell et al., 2008].
 The mean composition of aircraft samples collected over Niger during AMMA SOP0/DABEX is 20% (±2) for Al, 51 (±2) for Si, 11.6% (±0.6) for Fe, 9.1% (±0.5) for Ca, 3.6% (±0.2) for K, 1.6% (±0.2) for Ti, 2.2% (±0.2) for Mg, and 1.0% (±0.4) for Na. This is rather consistent with the mean composition obtained for advected dust from the ground-based samples. However, aircraft samples have less Si and Mg and more Ca, Ti, and Fe than the ground-based ones (t-test, 95% significance level). Artifacts, such as different sampling efficiencies of the aircraft and ground-based sampling inlets or efficiencies of analytical techniques, might provoke these differences. However, we do not expect them to be major. In fact we compared the number size distributions measured on the aircraft and at the ground during aircraft overpasses at low altitude (300 m above ground level) and found that is compared remarkably well over the region of overlap (0.3–3 μm diameter) of the particle counters operated airborne and ground-based. As it will be shown in section 3.3, differences in composition between ground-based and aircraft samples correspond to real mineralogical differences in the sampled dust. As a matter of fact, the aircraft did not fly during any of the episodes detected at Banizoumbou, except on 1 February, when the dust composition from ground-based and aircraft sampling is consistent.
 The DODO sample composition was similar than that of AMMA SOP0/DABEX. The exception was the mean composition of the dust storm which was encountered during flight B238 (23 August 2006). The mean percent composition obtained from three samples is 19% (±1) for Al, 46% (±1) for Si, 12.7% (±0.6) for Fe, 11% (±2) for Ca, 5.3% (±0.1) for K, 1.6% (±0.1) for Ti, 3.4% (±0.4) for Mg, and 1.4% (±0.1) for Na.
 The dispersion in the Ca values might reflect the contribution of air masses from different source regions that appeared to be at different heights [McConnell et al., 2008]. Differences with respect to the mean AMMA SOP0/DABEX dust composition (excluding the Bodélé sample) are observed for Si, Fe, Ca, Mg, and, for the first time, for K. This is consistent with the enrichment in K-bearing illite clays with respect to kaolinite expected between western Africa and the Sahel [Caquineau et al., 2002], which also became apparent from the XRD analysis (see section 3.3). Electron microscope observations also indicated that dust aerosols collected during B238 flight (and during the DODO experiment in general) were often complex aggregates containing K, whereas were mostly simple alumino-silicates made of Al, Si and Fe over Niger [Chou et al., 2008]. The high sample-to-sample variability prevents any conclusions regarding a possible enrichment in Na. This was observed during previous sampling in the region [Formenti et al., 2003], either because of mixing with sea breeze or as mineral halite (NaCl), which is one of the soil constituents of the local source regions, often dried saline lake deposits. Na was often observed by electron microscopy in many of the particles investigated.
3.3. Mineralogical Composition of Mineral Dust
 These results are in agreement with the observations by X-ray diffraction analysis (XRD), which could be performed on almost all the samples discussed in the previous section. For each sample, results are shown in Figure 2 (top) in terms of the percent of total diffracted peak surface of the XRD spectra due to the different minerals in the crystallized phase. Figure 2 (bottom) shows the ratio of the percent total diffracted surface of illite and kaolinite. Caquineau et al.  has shown that this ratio is sensitive to the source of provenance of African mineral dust.
Figure 2. (top) Percent of the total diffracted peak surface obtained by X-ray diffraction (XRD) analysis for various minerals detected in ground-based and aircraft samples and (bottom) corresponding illite-to-kaolinite ratios.
Download figure to PowerPoint
 There is some consistency in the mineralogical signature from sample to sample. Clays (kaolinite and illite) accounted for 50–60% of the measured signal, quartz for about 30–40%, K-feldspar and plagioclase (feldspar containing Na and Ca) for about 5%, and carbonate (calcite, dolomite and gypsum) for the remaining 10–15%. The largest fluctuations are observed again on the carbonate fraction, calcite in particular, which is greatly enriched (up to 20–30% of the signal) on some of the samples collected during AMMA SOP0/DABEX. The simultaneous detection of palygorskite at trace level (not shown) suggests that dust likely originated from North Africa (east of Egypt or Algeria). The dust collected during DODO2 (flight B238) has a rather similar composition. However, more subtle differences become evident when looking at the illite/kaolinite ratio. Dust collected over Niger (samples collected at the ground and during AMMA SOP0/DABEX) are characterized by an illite/kaolinite ratio of the order of 0.1–0.3, dust collected over Mauritania appears to be richer in illite (ratio of the order of 0.6). This is consistent with previous findings by Caquineau et al. .
 We must warn the reader on the fact that the percent of total diffracted surface cannot be used readily to infer the mass concentrations of the various minerals. First of all, not all the collected aerosol mass is analyzed. In fact, in order to reduce the detection limit of the technique, the aerosol deposited on the filter needs to be placed on a specific sample holder. To do so, a fraction of the original sample is used, which, in spite of the degree of precision in manipulation, may not be identical from sample to sample. Second, the XRD signal (the total diffracted peak surface) is due to the crystalline fraction only of the aerosol, and not to the total dust mass. This fraction might vary from sample to sample, as it is illustrated in Figure 3 by the scatterplot of the total diffracted peak surface of the ground-based samples versus the analyzed mass. The correlation is satisfactory (R2 = 0.64); however, the large scatter indicates that the fraction of diffracting material is not strictly equal from sample to sample. As an example, the total diffracted peak surface for sample SOP0-4 (circle) is lower by a factor of 3 than that of sample SOP0-29 (triangle), in spite of the fact that the analyzed mass is the same. Sample SOP0-4 was collected during the Bodélé outbreak and contained a large percent of amorphous diatoms debris.
Figure 3. Scatterplot of the total diffracted peak surface obtained by X-ray diffraction (XRD) analysis of the ground-based samples versus the aerosol mass analyzed by the technique. The total diffracted peak surface represents the total XRD signal provided by the fraction of the aerosols which is in the crystalline form. The total diffracted peak surface for sample SOP0-4 (circle), collected during the Bodélé outbreak of 15 January 2006, is lower by a factor of 3 than that of sample SOP0-29 (triangle), in spite of the fact of that the analyzed mass is the same.
Download figure to PowerPoint
 For a given mineral, the presence of amorphous phases (e.g., diatoms debris) prevents us from quantifying the crystalline fraction from the mass elemental concentration measured by XRF. Another difficulty is due to the fact that the same element might be present in various minerals. This is the case for quartz, which does not correlate with the silicon (percent of total oxide mass from elemental analysis). The clay fraction (illite + kaolinite) fraction did not correlate with the aluminum fraction either, likely because the Al content of each of these different species is different. Conversely, good correlation (R2 = 0.63) is found between the crystalline calcium fraction (calcite + dolomite) and elemental calcium, suggesting that all the elemental calcium is in the crystalline form (Figure 4).
Figure 4. Scatterplot of the crystalline calcium fraction (calcite + dolomite) and the Ca mass fraction (percent of the total mass estimated from elemental analysis), suggesting that all the elemental calcium is in the crystalline form.
Download figure to PowerPoint
 Finally, the iron-oxide content and mineralogy (speciation hematite and goethite) was investigated. The iron-oxide content could be measured on fourteen ground-based samples, three collected during episodes of advection and one during an episode of local emission. The mean ratio of iron oxide-to-total elemental iron was 59% (±5), and varied between 52% and 63%. The lowest value was found for sample SOP0-4 corresponding to the Bodélé dust outbreak, while the highest value corresponds to the episode of local production (SOP0-46). Our values are in agreement with those reported for Niger dust by Lafon et al.  for advection (43% (±10)) and dust from local erosion (65% (±5)). The iron-oxide content varied between 2.4% and 4.5% of the total estimated dust mass (expressed as sum of oxides); again the maximum and minimum values corresponding to dust from Bodélé and dust emitted by local erosion. These values agree with those of Lafon et al. , who found 2.8% (±0.8) and 5.0% (±0.4) for advected and locally emitted dust, respectively. Some differences might be due to the fact that we did not estimated the total dust mass in the same way.
 Six aircraft samples could be analyzed, five for dust over Niger and one for dust over Mauritania. Results are consistent with those obtained on the ground. The iron oxide-to-Fe ratio varied between 46 and 59%. The lowest iron oxide-to-Fe ratio is obtained for dust originated from northern Africa, whereas the highest value is obtained for sample B238 collected over Mauritania, where iron mining is extensive. This information is resumed in Figure 5 showing the iron oxide-to-Fe ratio as a function of the percent content of total iron in the total estimated dust mass (as sum of oxides).
Figure 5. Scatterplot of the iron oxide-to-Fe ratio and the percent content of Fe in the total estimated dust mass for samples collected at Banizoumbou (diamonds) and on board the BAe-146 during AMMA SOP0/DABEX, DODO1, and DODO2 (circles). Samples with contrasted iron oxide-to-Fe ratio are highlighted.
Download figure to PowerPoint
 Finally, spectral diffuse reflectance analysis performed on those samples indicated that iron oxides were composed by relative proportions of hematite and goethite in relatively constant proportions (30% and 70%, respectively). This is in accordance with previous findings [Lafon et al., 2006]. The Bodélé sample had no measureable signal. The iron oxide depletion in Bodélé dust is apparent.