Physico-chemical and optical properties of Sahelian and Saharan mineral dust: in situ measurements during the GERBILS campaign



This paper presents new results on the composition, size and shape of mineral dust particles from African sources which were obtained by analysis of bulk filter samples collected in June 2007 onboard the BAe-146 research aircraft of the Facility for Airborne Atmospheric Measurements (FAAM). The aircraft was operated over Mauritania, Mali and Niger during the Geostationary Earth Radiation Budget Intercomparisons of Longwave and Shortwave radiation (GERBILS) campaign. Dust sampled during the campaign originated from various sources, including locally in the Sahel as a result of large-scale convective activity.

Regardless of origin, clays (illite, kaolinite) dominated the total volume (79–90%); the remainder was composed of quartz, calcium-rich minerals (calcite, dolomite, gypsum) and alkali feldspars. Iron oxides, measured using a selective chemical extraction method, accounted for 1–3% of the total dust mass. The dependence of particle number size and shape distribution on the origin of dust seems minor too, although our results might be slightly misleading due to the fact that those kinds of data have been gathered on flights when dust had comparable origins and residence time.

Mineral dust is only weakly absorbing in the mid-visible wavelengths (single scattering albedo ω0 > 0.95 at 550 nm), and ω0 measured values can be reproduced by measuring the bulk fractions of the major minerals, i.e. clays, quartz, calcite and iron oxides. At this wavelength, knowledge of the nature of clays and iron oxides, or the state of mixing of the minerals, does not induce significant differences in the results. A more precise description of the nature of clays and iron oxides is necessary at lower wavelengths owing to larger differences in their spectral optical properties. In particular, knowledge of the nature of the dominant clay is important for determining light scattering in the backward hemisphere. Copyright © 2011 Royal Meteorological Society and British Crown Copyright, the Met Office

1. Introduction

By scattering and absorbing solar and terrestrial radiation, mineral dust has a direct impact on climate (Sokolik et al., 2001). At a global scale, the direct radiative effect of mineral dust from natural and anthropogenic sources at the top of the atmosphere (TOA) is between −0.56 and +0.1 W m−2 (Forster et al., 2007). On a regional scale, the radiative impact of dust can be much stronger. For instance, above the West African coast, Haywood et al. (2003) measured a reduction of direct radiation at TOA of −130 W m−2 in the solar spectrum during an intense episode of dust transport (aerosol optical depth ∼1.5 at mid-visible wavelengths).

Nonetheless, the direct radiative impact of mineral dust is still uncertain, owing to difficulties in estimating the concentration fields and the optical properties (single scattering albedo ω0, optical depth τ and asymmetry factor g), along with their variability in space and evolution with time (Forster et al., 2007). Direct measurements are only partially informative of the optical properties, which are not unique to an aerosol type. As a consequence, the prediction of the direct radiative impact requires a model enabling the calculation of the dust optical properties based on its intrinsic physicochemical properties (size, shape and mineralogical composition). This approach should also help in estimating the spectral variability of the dust optical properties, which are normally only measured at a few wavelengths in the visible.

Mineral dust is composed of quartz, clays (illite, kaolinite, smectite, chlorite), carbonates (calcite, dolomite), sulphates (gypsum), feldspars (albite, orthoclase, anorthite) and iron oxides (haematite, goethite) having spectrally different optical properties (Sokolik and Toon, 1999). The impact of these minerals on radiation depends not only on their nature but also on their mixing state. Sokolik and Toon (1999) demonstrated that for a given composition and fixed atmospheric conditions mixtures containing particles in the form of iron oxides and clay aggregates could enhance the light absorption properties of mineral dust in the solar spectrum. Otto et al. (2007) underlined the importance of the coarse particle fraction (particles of diameter Dp > 3 µm) as well as the ratio between the fine and the coarse particle fraction on the optical properties (ω0, g). Dust particles are not spherical but generally have complex and irregular shapes (Reid et al., 2003; Kandler et al., 2007; Chou et al., 2008). Recent studies tend to show that the particle asphericity only induces differences in the order of few percent on ω0 and g values (Mishchenko et al., 1997; Yang et al., 2007; Otto et al., 2009). Nonetheless, these weak differences can lead to additional cooling of the earth–atmosphere system in the solar spectrum compared to the case where particles are spherical, because of an enhancement of backscattering by the non-spherical mineral dust (Kahnert and Kylling, 2004; Kahnert et al., 2005; Otto et al., 2009).

Over the last few years, many ground-based and airborne field campaigns have been conducted in Africa, the most important dust source in the world (Prospero et al., 2002). These are the Saharan Dust Experiment (SHADE; Tanré et al., 2003), the Bodélé Experiment (BODEX; Todd et al., 2007); the AMMA Special Observing Period-0 Dust and Biomass-burning Experiment (AMMA SOP-0/DABEX; Haywood et al., 2008), the African Monsoon Multidisciplinary Analyses (AMMA; Haywood et al., 2008), the Dust Outflow and Deposition to the Ocean (DODO; McConnell et al., 2008) and the Saharan Mineral Dust Experiment (SAMUM-1; Heintzenberg, 2009), and the airborne campaign Geostationary Earth Radiation Budget experiment Intercomparison of Longwave and Shortwave radiation (GERBILS), which was conducted in June 2007 over Mauritania, Mali and Niger (Haywood et al., 2011).

The GERBILS campaign offered the chance of studying mineral dust in Sahelian Africa, south of 15°N. The Sahel is on the transport route of mineral dust emitted from the Sahara and transported by the African easterly jet towards the North Atlantic. Additionally, in summertime the Sahel is a source region of mineral dust emitted by soil erosion during the passage of mesoscale convective systems or by intense monsoon winds (Sow et al., 2009). These additional emissions occur on human-disturbed soils only (Rajot and Valentin, 2001) and therefore may be considered to be anthropogenic in origin.

This paper provides new data on the elemental and mineralogical composition, shape and number size distributions of mineral dust collected via in situ filter sampling and optical counting. The physicochemical data presented in this paper are then linked to the optical properties of mineral dust using a Mie-code for spherical particles (Bohren and Huffmann, 1983) and with a T-Matrix code for spheroid particles (Dubovik et al., 2006). In particular, we focus on the representation of the single scattering albedo ω0 with respect to that of the mineralogical composition. The representation of the phase function p(θ) and the asymmetry parameter g calculated in correspondence of different hypotheses of the mineralogical composition and the representation of the particle shape distribution are also discussed.

2. Experimental methods

Measurements have been carried out onboard the UK community Facility for Airborne Atmospheric Measurements (FAAM) BAe-146 research aircraft between Niger and the Mauritanian and Senegalese coasts. The BAe-146 can fly from around 15 m over the sea and from around 160 m over land, with a ceiling of about 12 km. The average flight endurance is 5 h.

2.1. Aerosol sampling

The aerosol sampling system onboard the BAe-146 consists of a thin-walled inlet nozzle with a curved leading edge (Andreae et al., 1988). Sampling is conducted under sub-isokinetic conditions. The air speed in the inlet is about 70 m s−1, whereas the free airstream speed of the aircraft is about 100–110 m s−1. This should lead to an overestimation of the coarse fraction relative to the fine fraction of the aerosol.

Aerosols were sampled by filtration onto two stacked-filter units (SFUs) mounted in parallel. Each SFU can hold a maximum of three filters on sequential 47 mm diameter polyethylene supports, but only one stage was used during GERBILS. Each SFU consisted of a Nuclepore filter of 47 mm diameter and nominal pore size of 0.4 µm.

Sampling was carried out during straight and levelled runs (SLR) at constant altitude. Sampling typically lasted between 13 and 50 min. The sampling duration depended on the aerosol load, which was estimated in real time based on concurrent online measurements, but also on practical factors such as the need to conciliate many different measurements (in situ, remote sensing) during the same flight. Blank samples were also collected. Field blank samples were manipulated as actual samples and exposed to the air stream for a few seconds. Additional laboratory blank samples consisted of filters taken out of the boxes and stored separately without any manipulation or exposure in between. They give information about contamination from the filters themselves. Both loaded and blank filters were stored in Petri dishes.

In total, 62 samples, including 12 blanks, were collected during GERBILS. Owing to an electronic problem of the mass flow meters, the air volumes could not be recorded. As a consequence, in the following we will express our results in terms of absolute mass (in ng) per sample and not, as commonly done, in terms of mass concentration.

A summary of information, including exposure periods and geographical coordinates, of the filter samples collected during GERBILS is shown in Table I.

Table I. Details of filter samples collected during GERBILS.
FlightDateSampleStart andLatitude (°N)Longitude (°E)Altitude (m)
number identifierend time (UTC)   
  1. Columns represent the flight number, date, of collection, sample identifier, start and end time of collection (UTC) and geographic position (latitude, longitude and altitude of the start and end point of collection).

B29519 June 2007B295_112:15:–12:35:18.0–18.011.1–9.64937–4934
B29621 June 2007B296_112:46:–13:03:13.1–13.410.7–12.13930–3950
B29722 June 2007B297_112:52:–13:42:13.4–16.018.4–16.91304–1321
B29924 June 2007B299_110:43:–11:13:17.9–18.00.9–2.71011–664
B30025 June 2007B300_111:15:–11:45:18.0–18.07.0–6.9709–712
B30127 June 2007B301_111:14:–11:43:17.9–18.03.5–5.1698–928
B30228 June 2007B302_111:53:–12:33:17.9–18.012.7–10.0669–695

2.2. Particle bulk composition

2.2.1. Total mass and elemental composition

The aerosol mass was determined gravimetrically on each sample by weighing filters before and after sampling. Weighing was performed at the Laboratoire d'Hygiène de la Ville de Paris (LHVP), France. Filters were weighed at controlled temperature and relative humidity (RH < 50%) with a Sartorius balance (M5P, absolute precision 1 µg). The error on the measured mass is estimated at 10 µg including the repetition variability.

Particle-induced X-ray emission (PIXE) technique was used to measure the elemental concentration of the following elements: Na, Mg, Al, Si, P, S, Cl, K, Ca, Ti, V, Cr, Mn, Fe, Ni, Cu, Zn, Br, Rb, Sr and Pb. This technique is very effective in the study of atmospheric aerosol elemental composition and in particular it is very sensitive in the quantification of crustal elements; moreover it does not require any sample pre-treatment, thus minimizing contamination risks. PIXE was performed using a proton beam at the 3 MV Tandetron accelerator of the LABEC laboratory of INFN (Florence), with an external beam set-up (Chiari et al., 2005; Calzolai et al., 2006).

Samples were irradiated for about 1000 s with a beam intensity ranging from 5 to 30 nA, depending on sample load, over a spot of ∼2 mm2. During irradiation, the filter was moved in front of the beam so that most of the area of deposit was analysed. PIXE spectra were fitted using the GUPIX code (Maxwell et al., 1995) and elemental concentrations were obtained via a calibration curve from a set of thin standards of known areal density within 5% (Micromatter Inc., Arlington, WA, USA). Detection limits ranged between 0.02 and 2 µg for elements from Zn to Na. Elemental masses on both laboratory and field blank filters were of the same order of magnitude, and lower than 4 µg. This suggests that sample handling did not induce any significant contamination on the samples. Finally, 50 out of the 62 collected samples were found above detection limits for every element of interest and retained for analysis.

2.2.2. Mineralogical composition

The identification of major minerals composing mineral dust (quartz, feldspars, clays, calcite, gypsum and dolomite), with the exception of iron oxides, was performed by X-ray diffraction (XRD) analysis at the Institut de Recherche pour le Développement (IRD) in Bondy, France. It should be noted that the XRD technique is based on the Bragg law relating the crystalline structure of a mineral to light diffraction. It does not allow quantifying the amorphous fraction which might also be present in mineral dust, organic or biological material as, for example, diatoms, which are found in some major source regions such as Bodélé (Chou et al., 2008).

The analytical procedure and semi-quantitative treatment are fully described by Caquineau et al. (1997), who adapted the sample preparation to low-mass mineral aerosol (load deposited on filter > 800 µg). Particles were first extracted from the filter with deionized water (pH ∼7.1), then concentrated by centrifugation (25000 rpm for 1 h) and finally deposited on a pure silicon slide (Queralt et al., 2001).

Analysis was performed using a Siemens D500 diffractometer with Ni-filtered Cu Kα radiation at 40 kV and 30 mA. Samples were scanned from 2° to 70° (2θ) with counting for 10 s every 0.02° 2θ. Owing to the mass limitation, only six samples could be analysed by XRD.

A calibration of the XRD spectrometer was performed in order to quantify the mineralogical composition. This consisted in establishing, for each mineral, a relationship between the intensity of the diffraction peak and the mass. The intensity of the diffraction peaks depends on the mass but also on the chemical composition and structural characteristics of the mineral. These are rather invariant for minerals such as quartz, carbonates and sulphates, but they can change dramatically for clay minerals depending on their origin and their ‘history’, i.e. the degree of alteration they have experienced since formation (Caquineau, 1997). As a result, laboratory standards might not be representative of the mineralogical state of clays in the aerosol samples. Therefore the clay content was established indirectly as the difference of the gravimetric and calibrated masses. This will be explained in detail in the following.

Calibration with pure standard minerals was performed for quartz (SiO2, Fontainebleau, France), calcite (CaCO3, Bédarieux, France), dolomite (CaMg(CO3)2, Traversella, Italy), gypsum (CaSO4, 2H2O, unknown origin) and orthoclase (KAlSi3O8, Madagascar). Their degree of purity was determined by XRD independently on the information that was available on the certification sheets. Binary mixtures of 2 mg of these standards (calcite with quartz, orthoclase with quartz, and dolomite with gypsum) in relative proportions varying from 0%, 20%, 40%, 60%, 80% and 100% were prepared. Similarly to the preparation of the filter samples, the binary mixtures were concentrated by centrifugation and then deposited on a pure silicon slide. In order to test the reproducibility of these measurements, each binary mixture was prepared and analysed three times. The measured XRD intensity and the deposited mass were found to be linearly correlated with a coefficient of linear regression R2 higher than 0.88 for quartz, calcite and gypsum (R2 = 0.88, 0.94 and 0.89, respectively) and equal to 0.55 for dolomite. The slope of the linear regression line was retained to convert the measured intensity into the mineral mass in the samples. Calibration of orthoclase was not possible since no linear relation was found between the intensity of the major peak and the deposited mass. A possible explanation for this is that preferential orientation of orthoclase particles occurred during deposition on silicon slides, particularly affecting the intensity of the major diffraction peak. Further experiments are in progress in order to overcome this difficulty.

Calibration data are given in Table II. As expected, the intercept was close to zero for all measurements. The variability of the slope of the mean regression curve has been estimated by taking into account the spread of the experimental points. This was found to remain lower than 21%, and mostly attributed to the variability in the sample masses due to weighing and manipulation.

Table II. Calibration data for minerals for which the calibration of the XRD spectrometer was performed (quartz, calcite, dolomite and gypsum).
MineralSlope (cps mg−1)y-intercept (cps)R2
  1. Columns represent the slope value obtained as the linear regression between the measured XRD intensity (counts per second (cps)) and the deposited mass (mg), percent standard deviation of the slope, y-intercept of the linear regression curve (counts per second), and the Pearson squared regression coefficient R2.

Quartz433 ± 560.2 ± 0.30.88
Calcite325 ± 290.9 ± 0.90.94
Dolomite679 ± 1450 ± 00.55
Gypsum446 ± 1100.4 ± 0.40.89

2.2.3. Iron oxide content

Iron oxide content relevant to light-absorption in the visible spectrum (Sokolik and Toon, 1999) was measured with the adapted CBD method developed by Lafon et al. (2004). This method is an adaptation for aerosol filters (mass smaller than 500 µg) of the classical method of Mehra and Jackson (1960) for soil analysis. The method uses the citrate, bicarbonate, dithionite (CBD) reagent to selectively dissolve iron oxides (Fe(ox)) via reduction. The remaining iron, called structural iron (Fe(struc)), occurs in the crystal lattice of silicates and is supposed not to contribute to the absorption of visible light (Faye, 1968; Karickhoff and Bailey, 1973).

The iron masses in the form of oxides mFe(ox) and that in the clay crystal lattice mFe(struc) are related by the following equation:

equation image(1)

where mFe(tot) is the total elemental iron mass. Before the CBD extraction, mFe(tot) is equal to the sum of the structural iron and iron oxide masses. After the selective extraction with CBD, mFe(tot) is equal to mFe(struc).

mFe(tot) was measured by wavelength-dispersive X-ray fluorescence spectrometry (WD-XRF). WD-XRF analyses were performed at Laboratoire Interuniversitaire des Systèmes Atmosphériques (LISA) using a PW-2404 spectrometer by Panalytical.

Excitation X-rays are produced by a Coolidge tube (maximum beam current 125 mA; maximum beam voltage 60 kV) with an Rh anode; the primary X-ray spectrum can be controlled by inserting filters (Al, at different thicknesses) between the anode and the sample. Each element was analysed three times, with specific conditions (voltage, tube filter, collimator, analysing crystal and detector), lasting 8–10 s. Data were collected using SuperQ software.

The elemental mass thickness (µg cm−2), i.e. the analysed elemental mass per unit surface, was obtained by comparing the filter yields with a sensitivity curve measured in the same geometry on a set of certified geo-standards (ANRT GS-N). These geo-standards were crunched in order to reduce and homogenize the grain size to particles smaller than 5 µm in diameter. They were then deposited with different concentrations (<150 µg cm−2) on Nuclepore filters. The atmospheric elemental concentrations were finally calculated by multiplying the analysed elemental mass thickness by the ratio between analysed and collection surfaces of each sample (16 and 28 mm, respectively).

2.2.4. Calculation of the mineralogical composition

A combination of the previously described measurements is used to quantify the dust mineralogical composition that is needed to estimate the complex refractive index. Two approaches are followed:

  • (a)For samples sufficiently loaded (mass > 800 µg) for which the mineralogical composition and the iron oxide content can be measured directly by XRD and CBD, the percentage composition by mass is established according to the following equation:
    equation image(2)
    where mtot represents the total mass, determined gravimetrically, mqz, mcal, mdol and mgyp represent the masses of quartz, calcite, dolomite, and gypsum obtained by calibration with reference standards and mFe(ox) represents the mass of iron oxides, respectively. Equation (2) is based on the assumption than the dust and the gravimetric masses are equivalent but also on the assumption that dust is entirely under crystalline form, and that the amorphous fraction (mostly silica in the form of diatoms debris) can be neglected. The validity of these assumptions for the GERBILS samples will be demonstrated in section 3.The mass of clay minerals, which cannot be determined via a direct calibration, can be obtained from Eq. (2) as the difference between the total gravimetric mass and the mass of quantified minerals. The quantitative partitioning of clays into its major species (illite, kaolinite, smectite) is not possible in the absence of proper reference standards. Nonetheless, this distinction is needed because of their different spectral optical properties (Sokolik and Toon, 1999). The apportionment of the total clay content into illite and kaolinite could be performed using the illite-to-kaolinite (I/K) mass ratios determined by Caquineau et al. (2002) for samples of equivalent origins. These authors showed that the I/K ratios vary between 0.1 for Sahelian dust to 2 for Saharan dust. In the framework of this paper, this apportionment is not done but the I/K ratio is left as a free parameter by assuming that clays were entirely either in the form of illite or in the form of kaolinite. The smectite content is neglected because of the impossibility of distinguishing it from the total clay content. In the same way, the partitioning of iron oxides into their major forms (haematite, Fe2O3 and goethite, FeOOH) is neglected and, in accordance with Lafon et al. (2006) and McConnell et al. (2010), the iron oxide content is either considered to be entirely in the form of haematite or in the form of goethite.
  • (b)For lightly loaded samples (mass ≤800 µg), as is the case most of the time for aircraft sampling, the mineralogical composition can be estimated indirectly by using the elemental concentrations measured by XRF or PIXE (Lafon et al., 2006; McConnell et al., 2010). In this approach, the dust mineralogical composition is approximated by four components, namely clays, quartz, calcite and iron oxides. The elemental concentrations of Al and Si are used to apportion clays and quartz, Ca to account for calcite and Fe(ox) and Fe ratio to account for the iron oxides content. See Lafon et al. (2006) for the explicit formulae. In this case too, clays and iron oxides were approximated by their pure forms (kaolinite, illite, haematite, goethite).

2.3. Size-segregated particle composition and morphology

The size-resolved morphology and composition of individual particles were investigated by energy-dispersive scanning electron microscopy (ED-SEM) at LISA using an instrument type JEOL 6301F equipped with an X-ray energy-dispersive spectrometer (Oxford Link Pentafet Detector and Link ISIS analyser, Oxford Instruments, UK).

Images of individual particles were acquired at various magnifications in order to investigate the largest possible size distribution. These have been analysed using the HISTOLAB counting program (Microvision Instruments, France). Both the acquisition and the analysis of the images were performed in a manual mode and were therefore rather time consuming. As a result, only a limited number of particles (<1400 per sample) could be analysed.

Based on the light contrast between the particles and the background filter, the particle projected area Aproj is measured as the number of contiguous pixels whose brightness is higher than a predetermined threshold value corresponding to the filter background. The projected area is then related to the geometric diameter Dp, i.e. the diameter of a spherical particle whose area is equal to the projected area.

To provide insight into the two-dimensional particle shape, the particle aspect ratio (AR) is calculated as follows:

equation image(3)

where Lproj is the major projected dimension. AR is equal to one for spherical particles and greater than 1 for elongated particles such as flattened (oblate) or elongated (prolate) spheroids, which are the simplest non-spherical shapes that can generalize the spherical shape. These are described in terms of volume axis ratio e = a/b, where a is the axis of spheroid rotational symmetry and b is the axis perpendicular to the axis of spheroid rotational symmetry. As a consequence, a sphere is a spheroid with axis ratio e = 1, an oblate spheroid is a spheroid with an axis ratio e > 1 and a prolate spheroid is a spheroid with an axis ratio e < 1. Therefore, the axis ratio e is equal to AR for oblate spheroids and to AR−1 for prolate spheroids.

2.4. Particle size distribution and optical properties

The aerosol number size distribution was measured with a Passive Cavity Aerosol Spectrometer Probe (PCASP-100X) for nominal particle diameters between 0.1 and 3 µm (15 size channels) and with a Small Ice Detector (SID-2) for particle diameters between 4.8 and 60 µm (26 size channels). The PCASP-100X measures the intensity of radiation scattered by particles between 35° and 120° at 633 nm, whereas SID-2 detects between 9° and 20° at 532 nm. Size-dependent corrections taking into account particle non-sphericity and refractive index were performed as described in Johnson and Osborne (2011).

A TSI 3563 three-wavelength integrating nephelometer measured the scattering (σs) and backscattering (σbs) coefficients at 450, 550 and 700 nm at 1 Hz time resolution. The nephelometer integrates the scattered light between 7° and 170° (total scattering) and between 90° and 170° (backscattering). Data were corrected for truncation and non-idealities according to Anderson and Ogren (1998).

A single-wavelength Radiance Research Particle Soot Absorption Photometer (PSAP) provided the absorption coefficient (σa) at 567 nm by measuring the light attenuation by particles deposited on quartz filters. As described in Johnson and Osborne (2011), the PSAP measurements were corrected according to the procedure of Bond et al. (1999).

As discussed in McConnell et al. (2008, 2010), both the nephelometer and the PSAP were operated in the cabin downstream of a Rosemount-type inlet. Because of the unknown passing efficiency of this inlet, the scattering and absorption coefficient measurements are considered to be representative of particles smaller than 3 µm in diameter and not of the full size distribution. This is an important limitation for mineral dust that has a significant coarse mode fraction.

The in situ measurements of σs and σa on SLRs at constant altitude were averaged over the exposure interval of the filter samples and used to calculate the single scattering albedo (ratio of the scattering coefficient to the extinction coefficient) ω0 at 550 nm. The spectral dependence of the absorption coefficient was considered to be negligible within its variability during SLRs.

2.5. Air mass dispersion and back-trajectory calculation

The origin of air masses encountered during the straight and level runs dedicated to filter sampling was determined by dispersion modelling.

This was undertaken using the Met Office Numerical Atmospheric-dispersion Modeling Environment (NAME). This is a Lagrangian particle model (Manning et al., 2003) in which emissions from pollutant sources are represented by parcels released into a model atmosphere driven by the meteorological fields from the Met Office's numerical weather prediction model, the Unified Model (Cullen, 1993).

The model output was generated using 100000 air parcels to represent the origin of the air arriving at the mean position of the aircraft (latitude, longitude and altitude) but the movement of the aircraft during the sampling is taken into account too (standard deviation of longitude and latitude). Output was restricted to those air parcels present within 500 m of the surface during transit. As such, it represents the source areas that are likely to have contributed to the dust aerosol observed by the plane.

2.6. Calculation of optical properties

The sensitivity of optical properties to the physicochemical properties of mineral dust was investigated using the code for randomly oriented oblate and prolate spheroids developed by Dubovik et al. (2006) with volume axis ratio e between 0.3 and 3. The code was used to compare the optical properties of spherical (e = 1) and non-spherical particles (e derived from experimental data as described in section 2.3).

Calculations were performed according to two mixing hypothesis. First, we considered that minerals are externally mixed with the exception of iron oxides, which we considered to be internally mixed with clays (Sokolik and Toon, 1999; Lafon et al., 2006).

In this hypothesis, the scattering and absorption coefficients σs and σa at a wavelength λ are obtained as the sum of the scattering and absorption coefficients of various minerals (indexed as i) as

equation image(4)

where Qs,i,j and Qa,i,j represent the single-particle scattering and absorption efficiencies for a particle of diameter Di,j and complex refractive index ñi,j and Ni(Di,j) represents the number size distribution of the i-mineral. Ni(Di,j) is related to fn,i, the number fraction of an individual mineral in the bulk dust aerosol.

The complex refractive index ñagg of the i-component corresponding to the aggregate between clays and iron oxides was estimated from the complex dielectric constant εagg of the randomly inhomogeneous binary mixture (i.e. without hypothesis on the form of the aggregate). This can be calculated using the Bruggeman approximation (Bohren and Huffmann, 1983) as follows:

equation image(5)

where v1 and ε1 are the volume fraction and the complex dielectric constant of the first component of the mixture, and ε2 is the complex dielectric constant of the second. The complex refractive index ñagg of the mixture (in our case, the iron oxide–clay aggregate) is related to the complex dielectric constant as follows:

equation image(6)

For the sake of comparison with previously published work, calculations were also performed assuming that all minerals are internally mixed. In this case, Eq. (4) is reduced to

equation image(7)

where ñj is the dust volume-averaged complex refractive index calculated using the classical volume-average mixing rule (Bohren and Huffmann, 1983) as follows:

equation image(8)

In this case fv,i is the volume fraction of the i-component and ñi its complex refractive index. For consistency with the externally mixed case, the Bruggeman mixing rule was still used to describe the iron oxide–clay aggregates.

The spectral complex refractive indices of individual minerals relevant to this study are listed in Table III.

Table III. Values of the complex refractive index of the major constituents of mineral dust at 450 and 550 nm used in this study (with their references).
Mineralnik, 450 nmnik, 550 nmReference
  • a

    Only a wavelength-independent mean refractive index value was considered between 0.4 and 0.7 µm.

Calcite1.58 − 0.057i1.58 − 0.057iQuerry et al. (1978)
Dolomite1.62 − 0.000i1.62 − 0.000iBarthelmy (2007)a
Goethite2.43 − 0.068i2.27 − 0.088iBedidi and Cervelle (1993)
Gypsum1.52 − 0.000i1.52 − 0.000iBarthelmy (2006)a
Haematite3.23 − 0.330i3.25 − 0.214iBedidi and Cervelle (1993)
Illite1.42 − 0.001i1.41 − 0.001iEgan and Hilgeman (1979)
Kaolinite1.49 − 0.000i1.49 − 0.000iEgan and Hilgeman (1979)
Quartz1.55 − 0.000i1.55 − 0.000iDeer et al. (1966)a

3. Results

3.1. Apportionment of the aerosol load and origin of mineral dust

Ten BAe-146 flights were carried out from 19 June to 29 June 2007 between Nouakchott (Mauritania) and Niamey (Niger) in cloudless conditions. A detailed summary of flight tracks is given by Johnson and Osborne (2011). Most of the flights were conducted over land, between 15° and 18°N. One flight (B297, 22 June) was performed over the ocean along the Mauritanian and Senegalese coasts in the outflow of a large dust storm, which had also been sampled the previous day (flight B296, 21 June). Flights consisted of vertical profiles and SLRs above, below and within the dust layers, in order to investigate the dust and the atmospheric vertical structure, the perturbation of the radiative fields due to the airborne dust, and finally the physicochemical and optical properties of mineral dust.

Examination of satellite images (e.g. Ozone Monitoring Instrument (OMI) extinction optical depth maps at 388 nm, not shown) suggested that dust was present and widespread during the entire experimental period, being emitted from source regions in Mauritania, Mali, Niger and south of Algeria, but also from the Bodélé depression in Chad.

The dominance of mineral dust to the aerosol load is confirmed by experimental data collected during the campaign. The range of variability of the total and elemental mass (in µg) is shown in Table IV. Elemental tracers of mineral dust, such as Al, Si, K, Ca, Ti and Fe, dominated the aerosol load, their sum in the oxide form accounting for up to 70% of the total gravimetric mass. However, if no correction is applied, Al and Si concentrations as determined by PIXE are surely underestimated because of self-attenuation of the emitted X-rays due to the particle size (Calzolai et al., 2010; Formenti et al., 2010); if this effect is properly taken into account, the contribution of mineral dust to the total aerosol load increases to approximately 87%. In addition, a contribution of marine aerosol up to 7% was evident on 27 samples collected close to the Atlantic coast by the linear correlation of Na and Cl (coefficient of linear correlation R2 = 0.82). This contribution is also likely to be underestimated as the signal from Na also suffers from self-attenuation. Very minor traces of Cu, Cr, Pb, Ni and Zn (representing less than 3% of the total gravimetric mass) were also detected on samples collected closely downwind of the Nouakchott area. The missing fraction is mostly attributed to carbon, which is not determined by X-ray analysis, but which is found in the aerosol composition and in that of mineral dust in particular. These conclusions are supported by the visual examination of individual particles by electron microscopy (not shown).

Table IV. Mean and standard deviation of the elemental masses (expressed in µg) measured during the GERBILS campaign.
ElementCampaign mean (± SD) (µg)
  1. SD, standard deviation.

Na5.5 (4.3)
Mg5.1 (4.3)
Al39.6 (44.0)
Si95.5 (108.0)
P0.46 (0.20)
S1.8 (1.0)
Cl4.2 (4.2)
K9.1 (9.6)
Ca14.2 (11.0)
Ti4.2 (4.6)
V0.25 (0.15)
Cr0.09 (0.10)
Mn0.57 (0.56)
Fe30.5 (32.2)
Ni0.02 (0.02)
Cu0.02 (0.02)
Zn0.06 (0.06)
Br0.02 (0.01)
Rb0.06 (0.07)
Sr0.22 (0.17)
Pb0.03 (0.02)
Total mass681 (709)

Examination of the vertical profiles of the spectral particle scattering coefficients indicated that dust layers extended from the surface up to 6 km (Johnson and Osborne, 2011). The vertical distribution was stratified, very intense dust plumes (scattering coefficient at times as high as 1000 Mm−1 at 550 nm) being found close to the surface but also between 3 and 5 km (scattering coefficient as high as 300 Mm−1 at 550 nm). The dispersion modelling results (shown in Figure 1) indicate that the Nouakchott and Niamey areas might have been under the influence of air masses coming from different sources. In addition to transport from eastern sources in the Saharan Air Layer above 1.5 km, which is common at this time of the year (Formenti et al., 2011) and which is evident in Figure 1 for the Niamey area, the Nouakchott area was most of the time also in the outflow region of dust plumes from north of Mauritania/Western Sahara/Morocco. This is also suggested by comparison of the vertical profiles of the scattering coefficients. An example at 550 nm is shown in Figure 2 for flight B299. A low-level dust plume was observed over Nouakchott but not over Niamey, whereas an elevated aerosol layer above approximately 2 km was evident over both locations, albeit with different vertical extent and stratification. The different origins of the surface and the elevated layers are suggested by back-trajectories (not shown), indicating a source in Mauritania and Western Sahara in the first case but easterly transport from Mali and North of Niger in the second.

Figure 1.

Surface air mass dispersion maps corresponding to filter sampling during the GERBILS flights. Colored areas represent geographical surface locations that have contributed to air masses encountered during sampling. The color scale is arbitrary, the degree of contribution increasing as the color becomes darker. The corresponding Fe/Ca ratio obtained by elemental analysis of each of the samples is indicated. This figure is available in colour online at

Figure 2.

Vertical profiles of the particle scattering coefficient at 550 nm measured during flight B299 over Niamey (left panel) and Nouakchott (right panel). The presence of additional surface transport in the vertical profile over Nouakchott is evident.

The dispersion model maps in Figure 1 are accompanied by a figure of the correspondent ratio of elemental Fe to elemental Ca (Fe/Ca ratio in the following), which is generally considered a good atmospheric indicator of dust origin (Kandler et al., 2007; Rajot et al., 2008; Formenti et al., 2008, 2011). Fe/Ca ratios below 2 are indicative of dust emitted from Saharan sources, whereas Fe/Ca ratios above 2 are associated with dust storms from the Sahelian part of West Africa. This simple analysis confirmed that dust sampled during flight B295 originated from a large storm that occurred over Mali as a consequence of a localized convectively driven cold pool (Marsham et al., 2008). Figure 1 also confirms the possibility of a west-to-east longitudinal gradient in dust origin, as on flight B301 (27 June), and to a minor extent on its reciprocal the next day (flight B302, 28 June). Finally, Figure 1 suggests that large areas might have contributed to the aerosol load collected during aircraft sampling, which occurs over up to 2° in latitude and/or longitude due to the fact that the aircraft covers large distances during sampling and that exposure times are large. The identification of isolated emission sources (often defined as emission ‘hot spots’) seems unlikely, at least without a robust way of identifying their activation at the exact time when the air masses had traversed the sources prior to sampling. As a consequence, aircraft samples should therefore be regarded as the integrative contribution of the ensemble of the dust ‘hot spots’ within the area indicated by the dispersion maps.

3.2. Characterization of mineral dust: physicochemical and optical properties

3.2.1. Mineralogical composition

Six GERBILS samples collected during flights B295, B300, B301 and B302 had masses larger than 800 µg and could be analysed by XRD. As suggested by Figure 1, the sample collected during flight B295 corresponded to dust emitted locally in the Sahel, whereas samples collected during flights B300, B301 and B302 corresponded to dust transported from the Western Sahara (Mauritania, Morocco and Algeria). The analysis shows some similarities in terms of identified minerals. As shown in Figure 3, these were clays (illite and kaolinite, very minor traces of smectite), quartz, feldspars, calcium carbonates (calcite and dolomite), and minor traces of calcium sulphates in the form of gypsum. In parallel, we also measured the percentage fraction of total iron in the form of oxides for the 12 GERBILS samples for which the collected mass was above 800 µg. With the exception of sample B300_1, five of those coincided with those analysed by XRD. The quantification of the mineralogical composition (by volume) obtained by combining these two types of analysis is shown in Table V. Regardless of the dust origin, clays, in the form of illite and kaolinite accounted for 79% and 90% of the total volume, respectively. The remaining fraction is composed of quartz (between 8% and 19%) and traces of calcite (≤2%). Gypsum and dolomite represented less than 1%. The iron oxides-to-total iron ratio for the analysed samples varied between 0.4 and 0.61, accounting for between 1% and 3% of the total gravimetric mass. According to Lafon et al. (2004, 2006), iron oxides were considered in internal mixing with clays. The volume fractions of the iron oxides into the clay–iron oxide aggregates are reported in Table V.

Figure 3.

X-ray diffraction spectra of the six GERBILS samples of masses larger than 800 µg. Spectra represent the number of X-rays as a function of the diffraction angle (expressed as 2θ). The major minerals identified are indicated by their first letter: S, smectite; I, illite; G, gypsum; K, kaolinite; Q, quartz; F, feldspar; C, calcite; D, dolomite.

Table V. Mineralogical composition (percent) by volume measured by X-ray diffraction and estimated using the measured elemental chemical composition for the different types of clay–iron oxides aggregates.
  Measured by XRDEstimated
 oxideoxide    oxideQuartzCalcite
 aggregate typeaggregate    aggregate  
  1. IH, illite and haematite; IG, illite and goethite; KH, kaolinite and haematite; KG, kaolinite and goethite. The volume fraction of iron oxide in the aggregate is indicated in parentheses in the columns labelled ‘Clay–iron oxide aggregate’.

B295_2IH90 (1.0)10<1 <1<175 (4.4)1115
 IG90 (1.2)10<1 <1<175 (5.3)1015
 KH90 (1.0)10<1 <1<154 (6.4)3016
 KG90 (1.2)10<1 <1<155 (7.7)3015
B301_1IH79 (0.8)191 <1<178 (4.3)16 7
 IG79 (0.9)191 <1<178 (5.2)15 7
 KH79 (0.8)191 <1<157 (6.3)36 7
 KG79 (0.9)191 <1<157 (7.6)36 7
B301_2IH90 (1.5)10<1 <1<177 (3.8)17 6
 IG90 (1.8)10<1 <1<177 (4.6)17 6
 KH90 (1.5)10<1 <1<156 (5.5)38 6
 KG90 (1.8)10<1 <1<156 (6.7)38 6
B302_1IH86 (1.2)112 <1<172 (3.3)1315
 IG86 (1.4)112 <1<172 (4.0)1315
 KH86 (1.2)112 <1<152 (4.9)3216
 KG86 (1.4)112 <1<152 (5.9)3216
B302_3IH89 (0.6) 91 <1<169 (4.5)1713
 IG89 (0.8) 81 <1<169 (5.5)1713
 KH89 (0.6) 81 <1<150 (6.5)3614
 KG89 (0.8) 81 <1<151 (7.9)3514

Table V also provides the percentage values of the mineralogical composition estimated using the elemental concentrations of Al, Si and Ca and those of the iron oxides. Large differences are evident. The impact of those on the calculation of the optical properties of scattering and absorption, and in particular on the single scattering albedo, will be discussed bellow.

3.2.2. Number size distribution

The normalized number size distributions (0.16 ≤ Dp ≤ 44.4 µm) measured during the eight SLRs corresponding to filter sampling during flights B299 to B302 are shown in Figure 4. There is a gap in the data between 3 and 6 µm in diameter as the PCASP and the SID size range do not overlap. This is potentially a large source of uncertainty in optical calculations as particles in this size range are very effective in interacting with radiation (contributing up to 20% of extinction and scattering and 30% of absorption). In the absence of further information, we used a linear interpolation to represent missing values.

Figure 4.

Upper panel: mean normalized number size distribution in accumulation mode (0.16 ≤ Dp ≤ 3 µm) and in coarse mode (6 ≤ Dp ≤ 44.4 µm) measured by the PCASP and the SID instruments, respectively, during the SLRs corresponding to filter sampling. Lower panel: an example of log-normal fit of the number size distribution measured during the SLR corresponding to sample B302_3.

In order to be used for optical calculations, the measured number size distributions were fitted with multi-mode log-normal functions. Since the aim is not to represent solely the microphysical modes but to describe as closely as possible the measured number size distribution for subsequent optical calculations, up to six modes were used in the parametrization. There is some inherent uncertainty in this procedure due to the high number of free parameters (18 degrees of freedom corresponding to the number of parameters of the log-normal functions) compared to that of points of data constraints (30 data points). This is true in particular for the first mode of particles smaller than 0.16 µm, which is not well constrained by measurements. Nonetheless this mode contributes little to scattering and absorption at the wavelengths considered in this paper (less than 1% at 450 nm) and its exact position is unnecessary. An example of log-normal fit is shown in Figure 4.

The summary of the modal log-normal parameters (median modal diameter Dmode, standard deviation σ and number fraction) used to parametrize the measured curves is shown in Table VI. The modal median diameter for particles of diameter smaller than 1.7 µm was found to be invariant from one sample to another. Conversely, their fractional contribution was variable. Some samples have a higher proportion of particles in the fraction below 0.4 µm and others in the fraction between 0.4 and 1.7 µm. Contrary to the fine particles, both the position of modal median diameter and the fractional contribution of particles larger then 1.7 µm in diameter varied. This variability seems to be related to the origin, or at least to the time after emission. There is a linear relationship between the fractional number of particles below 0.4 µm, between 0.4 and 1.7 µm and then that of particles larger than 1.7 µm in diameter and the Fe/Ca ratio measured in correspondence (not shown). Whereas the fractional number of particles below 0.4 µm is anti-correlated with the Fe/Ca ratio (slope of the linear regression line −0.8 ± 0.2; Pearson R2 coefficient 0.60), the fractional number of particles larger than 0.4 µm increases with increasing Fe/Ca ratio. In particular, the linear correlation is particularly satisfactory for particles larger than 1.7 µm (slope of the linear regression line 0.13 ± 0.02; Pearson R2 coefficient 0.87), suggesting that samples presenting a higher Fe/Ca ratio have been sampled more closely after emission, before the coarse fraction started to be depleted by gravitational settling.

Table VI. Summary of the modal log-normal parameters (mode diameter Dmode, standard deviation σ and number fraction) obtained for the multimodal log-normal deconvolution of the measured number size distribution.
SampleModeDmode (µm)σNumber fraction
B299_110.2 1.551.4%
 20.3 1.220.3%
 30.7 1.725.7%
 42.7 1.3 2.4%
 54.6 1.4 0.1%
 69.0 1.5 0.005%
B299_310.2 1.578.9%
 20.3 1.2 9.1%
 30.8 1.7 9.7%
 42.8 1.3 2.1%
 54.5 1.3 0.2%
 68.5 1.3 0.002%
B301_110.2 1.559.1%
 20.3 1.214.8%
 30.8 1.722.2%
 42.8 1.3 3.5%
 54.5 1.3 0.4%
 69.0 1.5 0.02%
B301_210.2 1.564.7%
 20.3 1.212.9%
 30.8 1.719.4%
 42.7 1.2 2.9%
 54.5 1.3 0.1%
 67.0 1.5 0.02%
B301_410.2 1.593.6%
 20.8 1.6 5.6%
 32.6 1.3 0.7%
 45.6 1.4 0.003%
B302_110.2 1.587.9%
 20.7 1.710.9%
 33.0 1.3 1.2%
 46.0 1.3 0.01%
B302_210.2 1.588.9%
 20.7 1.7 9.9%
 32.7 1.3 1.2%
 47.3 1.4 0.01%
B302_310.2 1.561.9%
 20.3 1.220.8%
 30.6 1.716.3%
 42.6 1.3 1.0%
 54.0 1.4 0.03%
 610.0 1.4 0.002%

The size-averaged number fractions fn,i of aluminosilicate, quartz and calcium-rich particles have been determined using SEM observations on two samples from flight B302 (samples B302_3 and B302_4). Results shown in Table VII confirm the dominance of aluminosilicate particles on the aerosol composition, accounting for up to 84% of the particle number. The aluminosilicate fraction is meant to represent the ensemble of the clay–iron oxide aggregates and feldspars. No isolated iron oxides particles have been observed, whereas iron was detected on practically all Al–Si-containing particles. The impact of minor differences in the number fractions on the calculation of the optical properties is considered bellow.

Table VII. Size-independent percent number fraction (fn,i) obtained by SEM for the major mineralogical components observed in the GERBILS dust samples.
  1. The analysis was performed for two samples B302_3 and B302_4 on 70 and 125 particles, respectively. Mean and standard deviation (SD) are also reported as a measure of the variability of the estimated number fraction.

B302 47810553
Mean (SD)81 (4)11 (1)4 (1)3 (2)1 (1)

3.2.3. Particle shape

The cumulative probability distributions of the particle AR for particles between 0.5 and 10 µm in diameter collected during flight B302 (samples B302_3 and B302_4) are shown in Figure 5. Particles smaller than 0.5 µm in diameter could not be imaged with sufficient precision to investigate their shape. Particles larger than 10 µm in diameter were too few to be represented with statistical significance. In addition, they were mostly found in the form of complex aggregates, difficult to attribute to a single particle or to an ensemble of particles, and the fractal form of which was difficult to resolve by the imaging program. Because the error induced by the manual description of their perimeter is high for these large and fractal particles, we chose to discard them from the analysis.

Figure 5.

Cumulative probability distributions of particle aspect ratio (AR) values obtained by SEM imaging for samples B302_3 and B302_4. Data have been grouped into four size classes of diameters ranging between 0.5 and 10 µm. The size-averaged distribution is also shown.

Data have been grouped in four size classes in order to increase the statistical significance. The AR value distributions varied between 1.1 and 5. Some size dependence seems to be evident for sample B302_4, when particles larger than 2 µm show a larger proportion of AR values greater than 1.75. The large counting uncertainties prevent results from being conclusive in this respect. A mean AR distribution independent of particle size has therefore been calculated with a median value of 1.55. These findings are in agreement with previous AR measurements on African dust particles (Kandler et al., 2007, 2009; Chou et al., 2008), suggesting that despite the physical diversity the particle shape distribution of mineral dust can be represented by a model distribution independent of particle size and origin.

In order to be used in optical calculations, the experimentally determined mean size-independent AR distributions were assumed to represent the shape distribution of an equal mixture of prolate and oblate spheroids. The optical code of Dubovik et al. (2006) is limited to AR < 3. The calculated axis ratio distribution (e = AR for oblate spheroid and e = AR−1 for prolate spheroid) extended from 0.3 to 0.9 and from 1.1 to 3.

In order to test the importance of values larger than 3, which represented around 9% of the observations, three extreme cases were considered: a first one in which the number of AR values > 3 was apportioned in equal parts to the extreme values of the e-distribution (e equal to 0.35 and e equal to 2.87, respectively); a second one in which the number of AR values > 3 was arbitrarily apportioned to the class of e-values equal to 0.35 (increased number of oblate spheroids); and a third one in which the number of AR values > 3 was arbitrarily apportioned to the class of e-values equal to 2.87 (increased number of prolate spheroids).

3.2.4. Scattering and absorption coefficients and single scattering albedo

The measured scattering and absorption coefficients σs and σa at 550 nm varied between 73 and 531 Mm−1 and 3 and 15 Mm−1, respectively. We recall that these values probably represent the contribution of particles smaller than 3 µm. The spectral dependence of the scattering coefficient σs between 450 and 700 nm remained weak (Angström coefficient < 0.2), as expected for mineral dust. In correspondence, the single scattering albedo ω0 varied from 0.91 (±0.02) to 0.97 (±0.01), with errors representing the combination of the experimental error and of the natural variability during each SLR. These measurements are comparable but on the lower edge of previous measurements performed at 550 nm during the SHADE, AMMA SOP0/DABEX and DODO field campaigns, also performed in the area under with the same experimental set-up and data treatment procedures (Haywood et al., 2003; Osborne et al., 2008; McConnell et al., 2008). The GERBILS values are close (although lower) to those of the summertime DODO campaign (DODO2 in McConnell et al., 2008) and to those of the AMMA wet season campaign conducted over Niger (Formenti et al., 2011). This could be imputed to the different mineralogy of the Sahelian soils, which are expected to be richer in iron oxides than Saharan soils (Claquin et al., 1999). Nonetheless, the iron oxide content alone cannot explain the variability of the single scattering albedo. This is made evident in Figure 6 representing the scatterplot of the ω0 values as a function of the percentage content of iron oxide in dust. This suggests that a simple linear parametrization of the single scattering albedo based on knowledge of the absorbing component of the aerosol is not possible and that more complete knowledge of the dust composition is needed as input to more rigorous modelling.

Figure 6.

Single scattering albedo averaged over the SLR corresponding to filter sampling as a function of the iron oxide fraction in GERBILS samples at 550 nm. Error bars represent the combination of experimental error and natural variability during each SLR.

The sensitivity of the single scattering albedo to the measured physicochemical properties is therefore investigated by calculation as described below. These have been carried out at 450, 550 and 700 nm for the four samples (B301_1, B301_2, B302_1 and B302_3) for which all the experimental data were available (mineralogical composition, size distribution, particle shape distribution, single scattering albedo). In order to compare the calculations with measurements, these have been performed first for particles smaller than 3 µm in diameter by limiting the number size distribution measured onboard the aircraft to this upper limit. Results of this investigation at 550 nm are summarized in Table VIII.

Table VIII. Measured and calculated single scattering albedo ω0 at 550 µm.
SampleMeasuredCalculated (external mixture, number fraction fn, i from sample B302_3)
  Measured mineralogical compositionEstimated mineralogical composition
B301_10.97 ± 0.010.960.960.980.980.910.910.900.88
B301_20.97 ± 0.010.950.950.970.960.920.910.910.89
B302_10.96 ± 0.010.960.960.980.970.940.930.930.92
B302_30.95 ± 0.010.970.970.980.980.930.920.920.90
SampleMeasuredCalculated (internal mixture)
  Measured mineralogical composition
  1. Calculations were performed for spherical particles (Mie code) on samples B301_1, B301_2, B302_1 and B302_3 for the different types of clay–iron oxides aggregates (IG, IH, KG, KH) using measured and/or estimated mineralogical composition. Either an external mineral mixture was considered and the number fraction fn,i measured from sample B302_3 (a) used, or an internal mixture was considered (b). Calculations are presented for particles smaller than 3 µm in diameter but are also indicated for the entire particle size range in parentheses.

B301_10.97 ± 0.010.970.960.980.97
B301_20.97 ± 0.010.960.960.970.96
B302_10.96 ± 0.010.960.950.970.96
B302_30.95 ± 0.010.970.970.980.98

3.3. Calculation of the optical properties

3.3.1. Dependence on the representation of the mineralogical composition

Calculations in spherical approximation have been performed in order to estimate the sensitivity of the single scattering albedo to the representation of the mineralogical composition, number size distribution and mixing state.

Two hypotheses have been tested: in the first one the clay–iron oxide aggregates (IH, IG, KH and KG) are externally mixed with the other minerals composing the dust samples (quartz, calcite, gypsum and dolomite); in the second one, the same clay–iron oxide aggregates are in internal mixing, according with the volume mixing rule with the remaining minerals.

In the external mixing hypothesis, calculations were also performed using the estimated mineralogical composition from elemental concentration measurements (Table V). It is intuitive that the large differences in the measured and estimated mineralogical composition observed in Table V must have an impact on the calculation of the single scattering albedo. The aim of this calculation is to quantify it and evaluate the pertinence of such an approximation. In both cases, we also tested the fractional contribution of each mineral compound to the total scattering and absorption by varying the mineral percentage number fraction fn,i according to the variability reported in Table VII.

Results are presented in Table VIII. The ω0 values calculated using the measured mineralogical composition are always very close to measurements, and independent of the type of aggregate. Because of the dominance of the clay fraction over the total mass, the nature of the iron oxide (haematite or goethite) and that of the dominant clay (illite or kaolinite) within an aggregate does not induce a significant variability in ω0. This is not the case when using the estimated mineralogical composition. In this case the measured ω0 is underestimated by calculations, and more dependent on the representation of the mineralogical composition. Underestimation of the ω0 values is related to the fact that the volume fractions of iron oxides in the clay–iron oxide aggregates and calcite, both absorbing in the mid-visible spectrum, are much higher when the mineralogical composition is estimated rather than measured (0.6–1.8% against 3.3–7.9% and 0.0–2.1% against 5.5–15.9%, respectively, for iron oxides and calcite). The variability of the calculated ω0 (estimated as the ratio of the standard deviation over the mean value) is doubled when the calculations are performed using the chemical rather than the mineralogical composition. This is because the volume of quartz and of iron oxide–clay aggregate fractions is more variable when they are estimated from the chemical composition measurements (the percentage standard deviation over the mean value are smaller than 96% and 19%, respectively) than when they are estimated using the mineralogical composition measurements (smaller than 0.6 and 0.08%, respectively).

At lower wavelengths, where the refractive indices of these minerals differ more significantly, the representation of the mineralogical composition could become more relevant. This is true in particular for the iron oxides, as the complex refractive index of haematite has a stronger spectral dependence than that of goethite (Bédidi and Cervelle, 1993). However, this is not really the case when the mineralogical composition is measured, as the spectral dependence of the refractive index of the iron oxides is smeared out by that of clays, which are rather spectrally neutral. At 450 nm, the calculated ω0 values in the external mixing hypothesis using the measured mineralogical composition are 0.95–0.97 for the IG aggregate, 0.93–0.96 for the IH aggregate, 0.96–0.98 for the KG aggregate and 0.94–0.97 for the KH aggregate. In this case, the smallest ω0 values are obtained when illite is mixed with haematite. The volumes of iron oxides fraction in the aggregates are the same for both kinds of aggregate but, at 450 nm, the imaginary part of the refractive index of illite is larger than that of kaolinite (Egan and Hilgeman, 1979). Furthermore, the spectral dependence for haematite-bearing aggregates is very small (ω0(450 nm)0(550 nm) ratio between 0.98 and 1).

At this same wavelength (450 nm), the ω0 values calculated by using the estimated mineralogical composition are always lower than when using the measured mineralogical composition (0.89–0.94 for the IG aggregate, 0.85–0.90 for the IH aggregate, 0.88–0.93 for the KG aggregate and 0.80–0.87 for the KH aggregate). In this case the lower ω0 is calculated from the mixture composed of the KH aggregate, because the haematite volume fraction of the aggregate is higher when the clay matrix is made out of kaolinite rather than illite (see Table V). This compensates for the fact that illite is more absorbent than kaolinite, which was the controlling factor in calculations performed with the measured mineralogical composition. When the mineralogical composition is estimated, the volume of iron oxide fraction in the aggregate is high and the influence of the nature and content of the iron oxide is clearly demonstrated. Finally, for the haematite-bearing aggregates, the spectral dependence is more significant (ω0(450 nm)0(550 nm) ratio between 0.90 and 0.97), owing to the higher volume fraction of iron oxides.

3.3.2. Dependence on representation of the mixing state, number size distribution and number fraction

Table VIII represents also the variability in ω0 at 550 nm, which is induced by varying the mineral mixing state. The comparison has been performed for the measured mineralogical composition by considering the case in which the clay–iron oxide aggregates, quartz and calcite are externally mixed (Table VIII(a)) and the case in which the clay–iron oxide aggregates, quartz and calcite are internally mixed according to the volume-averaged mixing rule (Table VIII(b)). The results of the comparison show that the variability induced in altering the state of mixing is minor and in any case always within the measurement uncertainties. This is also true at lower wavelengths (not shown), where, once more, the clay aggregates constitute the overwhelming fraction of the mineralogical composition.

As a consequence, this allows us to estimate a range of equivalent complex refractive indices which could be used by regional and global models to calculate the single scattering albedo of mineral dust. Depending on the type of aggregate, the real part of the refractive index is 1.44–1.52, 1.43–1.52 and 1.42–1.52, respectively, at 450, 550 and 700 nm. At the same wavelengths, the imaginary part is 0.001–0.0035, 0.001–0.0027 and 0.001–0.0032. The lower boundaries correspond to the GK aggregate and the higher boundaries to the IH aggregate. These values are in good agreement with recent estimates of the complex refractive index for African mineral dust (Osborne et al., 2008; Kandler et al., 2008; Müller et al., 2009; Otto et al., 2009; Petzold et al., 2009; McConnell et al., 2010).

Table VIII also shows that the variability induced by accounting for the fraction of coarse particles larger than 3 µm remains very small.

Finally, calculations have been performed assuming the mineral number fractions of samples B302_3 and B302_4.Varying these mineral number fractions (not shown) led to uncertainties in the ω0 calculations of the order of 1–3%.

3.3.3. Dependence on the representation of particle shape

The last part of the exercise consisted in estimating the influence of particle non-sphericity. This sensitivity test has been performed on two samples: sample B302_3, for which the axis ratio (e) distribution is known experimentally from the measured AR distributions but for which the calculations using the spherical model tend to overestimate the measurements (up to 3.8%); and B302_1, for which the axis ratio (e) distribution is not available by direct measurements but for which the calculation assuming spherical approximation was successful. This sample also had comparable size distribution to B302_3.

The exercise consisted in comparing Mie calculations for spherical particles (e distribution = 1 for all particles) to the random-spheroid calculations using the experimentally based size-independent e distributions (section 2.3.). Neither our shape analysis nor the random-spheroid software allows taking into account the possible dependence of the AR value distribution on composition, which nonetheless appears to be negligible for the minerals under consideration in this study (Kandler et al., 2008). Also, the size dependence of the refractive index was not taken into account. Calculations were performed assuming the internal mixing model for the mineralogical composition.

With these hypotheses, spherical/spheroid shape differences are found to be minor, remaining within 0.6% independent of the kind of aggregate and the wavelength, and also independent of the exact shape of the axis ratio distribution. Increasing the percentage fraction of oblate or prolate spheroid with AR > 3 did not modify the results. These results are in agreement with the work of Otto et al. (2009) showing that the spherical/non-spherical differences influence ω0 only within 1%.

The same aerosol models have been used to calculate the phase function and asymmetry factor g at 450 nm, shown in Figure 7 and Table IX, respectively, for sample B302_1. Besides the well-known differences between the angular scattering properties of spherical and non-spherical particles, these results highlight that the type of clay (illite or kaolinite) influences light scattering at angles higher than 90°. More precisely, the kaolinite aggregates scatter more light in the backward hemisphere than those composed by illite. These differences, which are important at angles above 120°, induce a variability of the order of 2–3% in the asymmetry factor g. For the single scattering albedo, the representation of values of AR larger than 3 does not influence the results of the calculations significantly.

Table IX. Calculated asymmetry factor at 450 nm for sample B302_1 using different mineralogical and shape-distribution models.
 Spherical modelSpheroidal modelMean ± SD
  Increasing e = 0.35IncreasingIncreasing 
  and e = 2.87e = 0.35e = 2.87 
  1. Mean and standard deviation (SD) corresponding either to the different representations of the particle mineralogy or to different representations of the particle shape distribution are shown. The percent SD is indicated in parentheses.

IG0.760.770.770.770.77 ± 0.003 (0.4%)
KG0.730.730.730.740.73 ± 0.002 (0.2%)
IH0.770.770.770.770.77 ± 0.004 (0.5%)
KH0.740.740.740.740.74 ± 0.002 (0.3%)
Mean ± SD0.75 ± 0.02 (2%)0.75 ± 0.02 (3%)0.75 ± 0.02 (3%)0.75 ± 0.02 (3%) 

4. Discussion and conclusions

Physicochemical and optical properties of West African mineral dust were measured in situ during the GERBILS aircraft campaign. Filter sampling was used to estimate the dust mineralogical composition. Results provide evidence that the mass content of iron oxides varied between 1% and 3% depending on the source region. As expected from knowledge of soil mineralogy, the highest iron oxide content was found in Sahelian dust emitted from Mali and Mauritania. These values are consistent with values obtained with identical sampling and analytical protocols during the AMMA SOP0/DABEX and DODO experiments (Formenti et al., 2008; McConnell et al., 2008, 2010), and in good agreement with values measured on dust generated in a wind tunnel from parent soils from Niger and Tunisia (Lafon et al., 2006).

Figure 7.

Phase function at 450 nm calculated for IG and KG mixtures and various hypotheses on particle shape distribution.

The mineralogical composition (clays, quartz, calcite, dolomite and gypsum) has been quantified for a limited number of high-loaded samples. The quantification, performed via calibration with reference standard minerals, reveals that clays account for between 80% and 90% of the dust composition, regardless of the dust origin. This estimate is in good agreement with previous results of Caquineau et al. (2002) for dust aerosols originating in the Sahelian belt. Our results also corroborate indirect evidence from ground-based remote sensing of the hyperspectral infrared downwelling radiance provided by Turner (2008) and suggest that the radiative signature can be described by considering that Sahelian mineral dust is composed predominantly of clay in the form of kaolinite.

Together with measured data on size and shape distribution, measurement of the mineralogical composition allows calculation of the single scattering albedo also measured onboard the aircraft at 550 nm. Owing to the dominance of clays over the dust mass, various hypotheses concerning the nature and mixing state of the minerals have been tested without finding any major impact on the calculations. This is also true for the iron oxide content, whose limited variability little influences the results, as well as for the type of iron oxide (haematite or goethite) considered in the calculations. Calculations performed at 450 nm, although without a possible comparison to measured values, indicate that in the near-UV the nature of the iron oxide gains in importance, owing to the larger differences in the imaginary parts of the refractive indices. Conversely, the computation of the phase function and asymmetry factor at this same wavelength indicate that the nature of the iron oxide is irrelevant in inducing variability, whereas the nature of the dominant clay is significant, particularly in the calculation of backscattering.

In conclusion, our results support the choice of Johnson and Osborne (2011) of modelling the integrated optical properties of mineral dust with a simplified model only considering haematite. Moreover, our results are in accordance with modelling estimates of Sokolik and Toon (1999) and Balkansky et al. (2007) suggesting that the absorption properties of mineral dust are weak and that the single scattering albedo of mineral dust can be modelled using an average iron oxide content of 1%. Furthermore, the independence of ω0 of the mixing type (internal vs. external) allows estimating an ‘effective’ mean refractive index which could be used in regional or global modelling of the direct radiative effect of the dust.

New data on the shape distribution of individual particles have been gathered. This part of the study agrees with previous a investigation indicating that the diversity of particle shape can be described by a unique parameter—the aspect ratio AR—whose distribution can be considered as independent of particle size. Dust particles are found to be non-spherical with AR values up to 5, which currently cannot be taken into account in optical modelling. However, a simple sensitivity calculation tends to suggest that strongly non-spherical particles (AR > 3) should not have a major impact on the calculation of optical properties.

To conclude, contrary to the shape distribution, investigation of the size dependence of the mineralogical composition should be undertaken to improve the results of this study and to help predict the time evolution of the single scattering albedo of mineral dust. Using our data, the lowering of the single scattering albedo by about 2% with the inclusion of the coarse mode is demonstrated (see Table VIII), but without taking into account the size-dependence of the complex refractive index.