We present aircraft measurements of dust aerosol during the Dust and Biomass-burning Experiment (DABEX), a project affiliated with the African Monsoon Multidisciplinary Analysis. DABEX took place between 13 January and 3 February 2006 in Sahelian west Africa, with the aircraft based at Niamey, Niger. The data set is augmented with Aerosol Robotic Network (AERONET) data. A mineral dust layer below 1–2 km (sourced from the north) and an overlying biomass burning (BB) layer (sourced from anthropogenic fires to the south) was observed on all days, although variability was observed in both layers. There is evidence of ozone loss within the dust, but with CO levels between 140 and 170 ppbv some history of combustion has occurred. Size distribution of the dust is compared with that of the BB aerosol and with dust measured near Senegal, during the Dust Outflow and Deposition to the Ocean (DODO-1) experiment. For accurate representation of the optical properties, five log-normals to the size distribution across sizes 0.05–5 μm are required, although two log-normals are adequate. The single scattering albedo was almost purely scattering, with values of 0.99 ± 0.01. During the strongest dust events the dust contribution to the column optical depth was 75–80%, compared to a DABEX mean of 50%. The aircraft-derived optical depth varied between 0.19 and 1.07, with the dust-only contribution between 0.07 and 0.81. AERONET optical depth trends are in good agreement with aircraft during DABEX, albeit with a bias to higher aircraft values. Retrieved AERONET aerosol size distributions show variable agreement with the aircraft. Differences between Versions 1 and 2 of the AERONET algorithm are highlighted.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 Airborne mineral dust particles have an effect on both regional and global climate through their interaction with solar and terrestrial radiation [Tegen et al., 1996]. They can also be chemically important as heterogeneous sites of destruction for trace gases such as ozone and oxides of nitrogen [de Reus et al., 2000]. Multidisciplinary campaigns involving research aircraft, satellites and ground-based monitoring have been conducted in recent years to assess regional impacts of mineral dust events. These include: the Aerosol Characterization Experiment ACE-Asia [Seinfeld et al., 2004], Saharan Dust Experiment SHADE [Tanré et al., 2003] and Puerto Rico Dust Experiment PRIDE [Reid et al., 2003], where the majority of measurements in these campaigns were carried out over oceanic areas. The ultimate aim of such studies, including the present one, is to better represent the aerosols in numerical weather forecast and climate prediction models. Since it is becoming clear that the physical, chemical and optical properties of dust varies from region to region, all the major arid regions of the world require scrutiny.
 The Sahara Desert is the single largest source of mineral dust aerosols in the world, with the Bodele Depression, Chad being the most intense region of deflation [Washington et al., 2003]. During the winter months (December–February), tropical central and west Africa is also a large source anthropogenic biomass burning (BB) aerosol which can interact with the mineral dust plumes in the sub-Saharan region. The state of particulate mixing, whether internal, external or some combination, can effect the size distribution and overall optical properties of the aerosol. In situ measurements of mineral dust particles show they can be internally mixed with sulphate, sea salt [Zhang et al., 2003] and black carbon [Clarke et al., 2004]. The mixing state of black carbon, present in BB aerosol, is of particular interest owing to its high shortwave absorption.
 In this study, aircraft measurements have been analyzed from the Dust and Biomass-burning Experiment (DABEX), part of a Special Observation Period (SOP-0) of the African Monsoon Multidisciplinary Analysis (AMMA). An overview of DABEX and SOP-0 is given by J. Haywood et al. (Overview of the Dust and Biomass-burning Experiment and African Multidisciplinary Monsoon Analysis Special Observation Period-0, submitted to Journal of Geophysical Research, 2008). The DABEX aircraft data set was complimented by ground-based in situ and remote sensing measurements that coincided with AERONET sun-photometer sites at Ilorin in Nigeria, Djougou in Benin [Mallet et al., 2008] and Banizoumbou in Niamey (J. L. Rajot et al., AMMA dust experiment: An overview of measurements perfomed during the dry season special observation period (SOP-0) at the Banizoumbou (Niger) supersite, submitted to Journal of Geophysical Research, 2008). In addition, this study complements measurements made with the same aircraft during the Dust Outflow and Deposition to the Ocean (DODO-1) experiment that took place off the coast of Senegal and Mauritania immediately after DABEX [McConnell et al., 2008]. Some of the major properties of Saharan mineral dust that advected over the Sahel region, i.e., the intrinsic microphysical, chemical and optical properties as well as geographic and vertical variation, are the subject of this paper. The interaction of dust with BB aerosol is assessed both in this paper and a related paper [Johnson et al., 2008a] focusing on aircraft observations.
 The Facility for Airborne Atmospheric Measurements (FAAM) BAe-146 research aircraft, used by the UK atmospheric research community, is equipped with state-of-the-art instrumentation that is described by Haywood et al. (submitted manuscript, 2008). However, the instruments central to this paper are described briefly below.
 The Passive Cavity Aerosol Spectrometer Probe 100-X (PCASP) is mounted externally under the aircraft wing and measures particles between 0.05 and 1.5 μm in radius. Owing to the short sampling lines of the PCASP and the relatively small maximum size it is assumed that the PCASP samples with 100% efficiency at all sizes. The sampling of supermicron particulates is problematic in aircraft sampling [Wendisch et al., 2004]. For particles larger than 1.5 μm, we rely on the use of a PCASP-X mounted inside the aircraft cabin that uses a counterflow virtual impactor (CVI) inlet operating in a passive aerosol mode [Johnson et al., 2000]. This instrument has an extended measuring range up to 5 μm radius. Although it is unlikely that 100% efficiency is achieved at these sizes owing to losses along the sampling line downstream of the inlet, quantifying such losses is difficult.
 The PCASP uses Mie scattering theory to derive a particle radius from the intensity of scattered radiation. Assumptions in deriving the PCASP size distribution include those associated with the shape of the aerosol particles (i.e., spheres), the refractive index (of calibration latex spheres, m ≈ 1.59-0.0i), and the dehydrating nature of the instrument caused by compressional heating and inherent warming from the electronics [Strapp et al., 1992]. Considering the moderate ambient humidities measured during the case studies in this paper (section 7), we assume the PCASP to measure fully dehydrated aerosol particles. With the mineral dust particles sampled being predominantly nonspherical [Chou et al., 2008], this raises the possibility of PCASP particle mis-sizing. Therefore it is prudent to assess errors in the assumption of spherical particles as is shown in section 5.
 The three-wavelength (0.45, 0.55, 0.70 μm) TSI model 3563 integrating nephelometer and single wavelength (0.565 μm) Radiance Research Particle Soot Absorption Photometer (PSAP) are used to measure the in situ scattering and absorption coefficients, respectively. They are mounted in the aircraft cabin and both use a Rosemount inlet. On the basis of previous work using Rosemount inlets, it is assumed that sampling efficiency drops off rapidly with particle size above about 1 μm radius [Haywood et al., 2003a]. Therefore we use the Anderson and Ogren  submicron correction coefficients for angular truncation. The “missing” aerosol scattering that is required for accurately assessing the column optical depth from aircraft profiles is estimated on the basis of the contribution implied from the CVI PCASP-X measurements of the coarse mode. The PSAP data were corrected for filter spot size, flow rate and overreading of absorption following the method of Bond et al. .
 AERONET aerosol optical depths (at 1.02, 0.87, 0.675 and 0.44 μm), single scattering albedos and aerosol size distributions from the Banizoumbou ground site are used in this paper. The latter two properties are derived during the inversion of scattered sky radiances measured during cloud-free Almucantor scans. Accurate absorption (i.e., single scattering albedo errors of about 0.03) is achieved when the aerosol optical depth at 0.44 μm is greater than 0.5 and the solar zenith angle is greater than 50° [Dubovik et al., 2002]. The AERONET sun photometry retrievals were carried out using both Version 1 [Dubovik and King, 2000] and Version 2 of the inversion algorithm and all data were subjected to the level 1.5 criteria. Version 2 is considered a marked improvement over Version 1 because of (1) the treatment of surface albedo based on reflectance climatologies [Sinyuk et al., 2007] leading to more accurate retrieval of absorption, and (2) the refined treatment of nonspherical particles leading to more accurate size distribution retrievals [Dubovik et al., 2006]. In addition to the AERONET stations, a hand-held Solar Light Company MICROTOPS II 5-channel sun-photometer was used throughout most days to measure the optical depth in central Niamey.
 An alternative method for determining the coarse mode (as well as the accumulation mode) is from the AERONET inversion algorithm. Errors in the AERONET retrieval become relatively large for coarse mode particles owing to the relatively small contribution they tend to make to the optical depth and also because of the limitation of retrieving information on 5–10 μm particles with the longest AERONET wavelength of 1.25 μm. Because the CVI PCASP-X provides data more compatible with the other aircraft instrumentation, this instrument is used to derive the coarse mode size distribution with the caveat that considerable uncertainties exist.
 DABEX involved 14 flights of the BAe-146 aircraft between 13 January to 3 February 2006, based out of Niamey Airport, Niger (Haywood et al., submitted manuscript, 2008). Aircraft manoevres consisted of deep profiles from near the surface to above the BB plume at 4–5 km altitude. Straight and level runs (SLRs) were performed at various altitudes within the different aerosol layers. Given that aircraft profiles cover a far greater distance horizontally than vertically (the aircraft flies at nominally 110 m s−1 and climbs/descends at 5 m s−1), it is important to bear in mind variability in the horizontal by using SLR data to back up profile data.
 Mineral dust aerosol was encountered during all the DABEX flights, whether present in the elevated aged BB plume over the region, or mixed within fresh BB plumes to the south of Niamey, or within low-level “pure” dust plumes that were devoid of BB aerosol. The source of the dust in all these environments was predominantly regions of the Sahara Desert to the north of Niamey; see Haywood et al. (submitted manuscript, 2008) for a schematic of the synoptic-scale aerosol dynamics during DABEX. On three flights during DABEX (B160 on 21 January, B161 on 23 January, and B165 on 30 January 2006) the aircraft was able to sample dust plumes located to the northeast of Niamey, i.e., away from the possible influence of industrial sources in the Niamey conurbation. Figure 1 shows the track of the BAe-146 aircraft for these three flights, where the flying around Niamey was at various altitudes between 0 and 5 km and the flying to the northeast tended to be at low level (below ∼2 km) within the dust plumes. Although these plumes reached the Niamey region itself, they were relatively weak compared to the measurements made to the northeast. That said, this low-level dust plume was present on all flying days during DABEX and the peak value of the extinction coefficient within the total aerosol column was greatest within this dust layer on all flights except B159, detailed by Johnson et al. [2008a]. Mineral dust at low level was also found externally mixed in significant loadings within fresh BB plumes to the south in Benin and Nigeria; Johnson et al. [2008a] describes such interaction of BB and dust in more detail.
4. Vertical Distribution
 A low-level mineral dust layer existed below 1–2 km over a considerable geographical region throughout DABEX from north of Niamey to south within the BB source regions, although the magnitude of the dust loading varied considerably from day to day. This variability is captured by the high-resolution aircraft data in deep profiles that spanned both the dust and BB layers. There were deep profiles within the Niamey region on all flying days, so we have reasonable statistics for inferring typical vertical profiles [Johnson et al., 2008b]. Figure 2 shows four profiles of the in situ aerosol extinction coefficient (σextλ) close to Niamey on different days. Mean and standard deviation from SLRs have also been plotted to show variability at different altitudes. The extinction within the BB aerosol layer aloft, generally above ∼2.5 km, is relatively stable between the four days with values of σext0.55μm at about 0.1 km−1. The magnitude and depth of the elevated BB aerosol layer encountered to the northeast of Niamey around 18°N during the dust-searching flights (Figure 1) was similar to that found near Niamey. The spread in the three wavelengths, indicative of the Angstrom exponent, α, and hence particle size, is quite variable. B165 shows the smallest sensitivity to wavelength (small α) in the BB aerosol layer due to a bias to large particles caused by dust externally mixed with the BB aerosol. B161 shows the largest sensitivity to wavelength (larger α) owing to a bias to smaller particles and hence less externally mixed dust. The lowest values of α are within the dust layers where the spread of σextλ is small and the 0.55 μm wavelength is scattered to a similar or greater extent than the 0.45 μm wavelength. The distribution of dust in the vertical varies from a pronounced peak at ∼1 km in B160 and B161, to well-mixed in B165, to a highly variable layer 2 km deep in B164 (on 28 January 2006). In most cases, the dust plume merged into the BB plume with little or no separation, but on occasion (e.g., B161) a distinct, thermodynamically stable “clear slot” of high visibility was observed.
4.2. Trace Gases
 Carbon monoxide is a good tracer of fossil fuel and biomass combustion due to its long lifetime of several weeks in the atmosphere [Parish et al., 1993]. Ozone is produced as a secondary product involving NOx and volatile organic carbon emissions in BB plumes and has been shown in the past to be destroyed on the surface of mineral dust particles [Hanisch and Crowley, 2003]. Profiles of CO and ozone are shown in Figure 3 that are from the same profiles used in Figure 2. There is a distinct correlation between the CO, ozone and aerosol extinction in aged BB plumes. CO mixing ratios within the clear slot of B161 drop to about 100 ppbv which is considered slightly elevated compared to pristine background air. Within the dust plumes, CO values are indicative of significantly polluted air, ranging from 140 to 175 ppbv even though there is no evidence from the aerosol measurements of BB or industrial aerosol based on the aerosol size distribution and the absorption coefficient from PSAP.
Figure 4 explores the relationship between CO and ozone from all available data from flights B160 and B161 where strong dust plumes to the northeast of Niamey and plumes of significant BB aerosol in the vicinity of Niamey were investigated. The scatterplot of 1 Hz data has been split on the basis of the aerosol Angstrom exponent from the nephelometer measurements into dust, BB dominated and “mixed” aerosol. The α > 0.7 points show a strong positive correlation within the BB plumes between the two trace gases but within the lower dust layers this relationship breaks down as shown by the α < 0.15 points, suggesting a possible degradation of ozone due to either its loss onto mineral dust surfaces (including the ground) or long timescale advection from other pollution sources. Timescale chemistry modeling is required to explain the ozone loss processes in these case studies further.
4.3. Trajectory of Airmass Origin
 The Met Office NAME (Numerical Atmospheric Dispersion Modeling Environment) model has been used to help identify the probable source regions of the dust and BB aerosol observed by the aircraft. NAME is a Lagrangian particle model [Manning et al., 2003] driven by synoptic meteorology from the Met Office Unified Model [Cullen, 1993]. Tens of thousands of particles were released over Niamey (13°29′N, 2°10′E) for two separate runs on 21 January 2006 (flight B160): at 0.5 km (±10 m) corresponding to the dust layer; and at 3 km (±1 km) corresponding to the BB layer. In this instance NAME was run in reverse to indicate the origin of the air over the previous 5 days and the data are shown in Figure 5. For the dust case the data show air masses sourced from a layer 0–0.5 km; for the BB case the data show air masses sourced from a layer 0–2 km. The plots of air mass orgin show probable sources of dust to the north and also to the east near Lake Chad and show sources of BB aerosol to the southeast of Niamey. The elevated levels of CO within the dust layer are difficult to explain, but Figure 5 shows that either European or BB sources could be the cause: the dust plumes contain highly aged polluted air that has lost all its pollution aerosol but maintains a significant CO signal owing to its much longer atmospheric lifetime. Air mass origin maps have been extended to 15 days (not shown) but the geographical spread in the data points becomes too large to be meaningful.
5. Assessment of Size Distribution and Optical Errors
 Before presenting measured size distributions, an assessment of the likely errors of using the PCASP to measure mineral dust size distributions is included here. T-matrix calculations [Mishchenko, 1991] for randomly orientated nonspherical particles have been performed to compare against Mie-Lorenz theory. Two shapes of nonspherical dust particles are chosen, and in addition spheres possessing refractive indices of both latex and dust are considered. The nonspherical particles are prolate spheroids and prolate cylinders and both have aspect ratios of 1.7, where this number was the median value found from single particle electron microscope analysis of BAe-146 filter samples [Chou et al., 2008].
 Normalized scattering phase functions (at 0.55 μm) of dust particles at four specific radii corresponding to PCASP midchannel sizes (0.055, 0.137, 0.52 and 1.62 μm) for spheres, prolate spheroids and prolate cylinders are shown in Figure 6. Estimation of the dust complex refractive index (1.53-0.0004i) is discussed at the start of section 7. The PCASP sensing collection angle, 35–120°, is annotated in each plot. The forward scatter peak sharpens and increases as the particle radius increases, as does the discrepancy in phase functions between the three particle shapes. The largest discrepancies in the phase functions are at large scattering angles (i.e., in reflection) but these scattering angles are not captured by the PCASP. The oscillations in the spherical phase function for larger radii show interference patterns, but if the phase function is integrated over a size distribution function, then these oscillations smooth out.
 Integration of the phase function over the PCASP collection angle and multiplying by the scattering cross section for all particle size bins produces the predicted PCASP response (at the laser wavelength of 0.635 μm) shown in Figure 7, where latex spheres have also been included. The y axis shows a huge range of values and so the differences between the distributions are significant. Table 1 lists the change in midbin radius between latex spheres and dust cylinders (the two shapes showing the largest discrepancy), and the contribution of each bin to the scattering for dust spheres when integrated across a typical DABEX dust size distribution (the contributions are similar for the other shapes). Table 1 shows there is a tendency to undersize most particles by up to about 20%. The first six size bins (up to bin size 0.137 μm), for all particle shapes, contribute only about 2% to the total scattering; the contribution increases rapidly from bin 7 and is significant up to bin 19 (2.85 μm).
Table 1. Effects of Nonsphericity and Refractive Index on the PCASP Midbin Radii and the Contribution to Scattering at 0.55 μm for Dust Spheresa
Bins 1–15 are for the wing-mounted PCASP; bins 16–21 are for the CVI PCASP-X.
 Integration over the phase function (35–120°) and over the typical dust size distribution function reveals that PCASP scattered light energy at midvisible wavelengths, relative to latex spheres, decreases for prolate cylinders by 19%, decreases for prolate cyclinders by 10%, and decreases for dust spheres by 5%. In applying Mie theory to obtain total optical properties (as we have done in section 7) from the measured size distribution, i.e., effectively integrating over the full phase function, there again exists an error due to particle shape. The maximum error (from cylinders) in the total scattering is about 21%, and likewise for the mass specific extinction; the errors in the asymmetry parameter and single scattering albedo are negligible (for the latter quantity the absorption is very small).
 The error in the assumption of spherical dust particles is moderate compared to the other errors inherent in the derivation of the total optical parameters. An error of 20% is estimated for the density of mineral dust, leading directly to a 20% error in the mass specific extinction. The error in the measured absorption coefficient, our constraint on the imaginary part of the refractive index, is at least 25% [Schmid et al., 2006], plus the fact that this measurement relates to the submicron fraction only and so we do not have absorption data for the coarse mode. In the analyses that follow we have used the latex calibration response for the PCASP (section 6) and derived Mie optical properties assuming dust spheres (section 7). Therefore the analyses are consistent with past work with the same probe and compare easily to other studies. The purpose of the T-matrix error analysis was not to “correct” the PCASP calibration, as the assumption of spheroids or cylinders is not realistic, but to estimate likely error bars.
6. Aerosol Size Distribution
6.1. Number Distribution
 Any airborne dust size distribution depends on multiple factors which include: parent soil type and its loose fraction, wind velocity, roughness length, and ageing through sedimentation [Grini and Zender, 2004]. Previous studies on mineral dust size distributions have often attributed three lognormal modes covering sub- and super-micron sizes of dust with the largest median radii up to 20 μm [e.g., d'Almeida, 1987; Alfaro and Gomes, 2001; Clarke et al., 2004]. Alternatively, the aircraft measurements of Haywood et al. [2003a] in Saharan dust showed that five log-normal modes were required to fit the data. The median radii, magnitudes and standard deviations of the modes vary from study to study owing to the time-varying nature of dust particulates and the source region; that is, there are no physical reasons why the component dust aerosol modes should be constant.
 The mean aerosol size distribution from the DABEX “pure” dust cases from the northeast of Niamey across the accumulation and coarse modes is shown in Figure 8, together with the mean DODO-1 dust size distribution in the coastal region of Senegal and Mauritania [McConnell et al., 2008], and the elevated BB size distribution from flight B160. Differences between the BB and dust concentrations are obvious at all particle sizes. The BB aerosol comprises a dust mode at sizes greater than ∼0.3 μm, albeit at much reduced concentrations [Johnson et al., 2008a]. The dust distribution from DABEX has a flatter appearence with little to discern between the accumulation and coarse modes as there is in the BB aerosol. Although the DODO-1 distribution shows a distinct accumulation mode that is quite different to that of the DABEX dust, there is no evidence to suggest that the DODO1 dust was mixed with another aerosol such as from BB [McConnell et al., 2008]. Differences between the shape of the submicron dust aerosol between DODO-1 and DABEX may reflect inherent microphysical properties of the parent mineral dust and the effect of wind speed on particle saltation. The shape of the dust supermicron dust distributions are similar, however. The first channel of the PCASP suffered from electronic noise during DODO-1 so the data have been removed in Figure 8.
 Log-normal fits have been applied to the mean DABEX dust data (Figure 9) and the optimal combined fit has been achieved by minimising errors in the distribution of Mie extinction with particle size, such as that shown in the right-hand column of Table 1. Like Haywood et al. [2003a], we found that five log-normals are required for an accurate fit to the observations. The submicron and supermicron fractions contribute about equal amounts to the DABEX dust total extinction, in contrast to the DODO-1 dust which has a more pronounced coarse mode with a 75% contribution to the extinction. In contrast to BB aerosol where a good fit is only required for submicron aerosols [Haywood et al., 2003b; Johnson et al., 2008a], here it is important to characterize the supermicron aerosols. Table 2 summarizes the log normal parameters for the five modes that are required to describe the generic DABEX dust size distribution in Figure 9 over the accumulation and coarse modes. Using log normals allows reproduction of these data for use in transport and radiation models. On the basis that each aerosol mode is governed by processes of generation and loss which are at least to some degree independent from one mode to the next, the use of five modes at radii of less than 2 μm is probably unphysical. Additionally, numerical weather prediction models require the number of aerosol modes to be transported be kept to a minimum [Greed et al., 2008]. Therefore, at the expense of some small accuracy in the optical extinction, the size distribution in Figure 9 has been fitted with two log-normal modes and the log normal parameters of this fit in terms of number and mass are shown in Table 3. A larger third mode, as observed and modeled in some studies, was not observed in the present work owing to the age of the aerosol and possible sampling issues on the aircraft.
Table 2. Log-Normal Parameters for the Generic DABEX Mineral Dust Number Size Distribution
Table 3. Reduced Two-Mode Log-Normal Parameters for the Generic DABEX Mineral Dust Size Distribution
6.2. Volume Distribution
 Size distributions are routinely retrieved from inversions of AERONET clear sky alcumantar scans of radiance as a function of scattering angle. Therefore we compare AERONET retrievals with aircraft in situ measurements that took place in the vicinity of AERONET sites. AERONET data have been selected according to the criteria stated by Dubovik et al. . Figure 10 shows comparisons of aircraft and AERONET volume size distributions from flights B160 and B165 for versions 1 and 2 of the AERONET algorithm. The AERONET site is Banizoumbou (see Figure 1) in both cases. These are both days where the dust layer dominated the aerosol column in terms of optical depth. The AERONET distributions are the optimal solution to the inversion of the measured sky radiances, and so represent a columnar distribution which is constant with height. Therefore, in order to compare like with like, the aircraft data are a column mean distribution weighted according to the optical depth contribution of each aerosol layer. All the distributions in Figure 10 have been normalized by dividing through by the peak in the distribution. Indeed one of the problems in comparing these distributions is one of normalisation. Figure 10 suggests that the agreement is poor in the accumulation mode and good in the coarse mode, but this could be an artefact of the normalization; we are comparing the relative strengths of the modes.
 Comparing Versions 1 and 2 of AERONET, the bimodal nature of coarse particles (at about 2 μm and 4 μm) is lost in going from version 1 to 2 in B160. This double peak in the coarse mode was seen on other days using version 1 and is presumably not real. There is a large difference in the shape of the distribution in the submicron particles in B160 between Versions 1 and 2. For B165, differences are less but there is a large peak in the smallest channels for Version 1 (a known artifact for AERONET retrievals), which is removed in Version 2. A strong bias of aerosol volume to the coarse mode can be seen in both cases in the aircraft data, with B165 hardly showing an accumulation mode at all. During B165, the comparison between aircraft and AERONET is reasonable although still with significant differences in the submicron region. For B160, the comparison is fairly poor with an order of magnitude separation in the accumulation mode. The accumulation and coarse mode radii on both days are biased to slightly smaller values for AERONET compared to the aircraft data, although the width of the coarse mode agrees with the aircraft data. Therefore, AERONET is not fully consistent with the aircraft data in its derivation of the full size distribution in conditions of multiple aerosol layers, i.e., dust aerosol beneath BB aerosol.
 Conditions of single aerosol layers, such as shown by Haywood et al. [2003c] for BB aerosol in southern Africa, have produced good agreement between AERONET (Version 1) and aircraft for submicron aerosol distributions. Problems in the coarse mode, however, were illustrated and attributed to refractive index assumptions for the PCASP analysis and the increase in error caused by trying to retrieve particle sizes much larger than the maximum wavelength of AERONET (1.02 μm). The present work also highlights problems of the effects of particle nonsphericity, which increases errors both for aircraft in situ measurements and remote sensing by AERONET, plus the possibility of large particle sampling inefficiencies on the aircraft.
7. Aerosol Optical Properties
 The pertinent optical properties at a wavelength of 0.55 μm of single scattering albedo (ω0), mass specific extinction (ke) and asymmetry parameter (g) for mineral dust measured during DABEX, DODO-1 and SHADE are summarized in Table 4. ω0 is shown both from observations and from Mie scattering calculations; ke and g are shown for Mie calculations only. The density of dust used to calculate ke is assumed to be 2.65 g cm−3 [Haywood et al., 2003a] across all particle sizes. The Mie calculations have been carried out using log-normal fits to the measured data as input, with the complex refractive index also shown in Table 4. The real part of the DABEX refractive index (1.53) was taken from WCP  and the imaginary part was adjusted until the computed value of ω0 equalled the mean value measured with the nephelometer and PSAP. The imaginary part of the refractive index for DABEX (0.0004) was similar to that for DODO-1 (0.0005); the higher value from SHADE (0.0015) reflects inherently more absorbing dust particles.
Table 4. Optical Properties of Mineral Dust Aerosol at 0.55 μm From Various Aircraft Compaignsa
keMie, m2 g−1
AM and CM refer to accumulation mode and coarse mode.
0.99 ± 0.02
1.53 + 0.0004i
1.53 + 0.0004i
AM + CM
0.99 ± 0.004
1.53 + 0.0005i
1.53 + 0.0005i
AM + CM
0.97 ± 0.02
1.53 + 0.015i
1.53 + 0.015i
AM + CM
 We can also use the aerosol mass spectrometer (AMS) data as described further by Capes et al.  to help us analyze aerosol chemical components. Using sulphate and total organic mass loadings as suitable indicators of industrial and BB aerosol that the AMS measures, the mass loading of these two components within dusty SLRs varied between 0 and 2% of the total submicron PCASP aerosol mass on the basis of data from B160, B161 and B165. This contrasts with an equivalent range of 24–46% within the elevated BB layers during the same flights. Therefore there is agreement between the AMS and other aerosol information such as the PSAP absorption coefficient, that what we are labeling as “pure” dust indeed contained negligible influence of BB or industrial aerosol.
 Comparisons of the optical depth, ω0 and α that are representative of the total aerosol column for specific aircraft profiles (the same profiles used in Figures 2 and 3) are made with AERONET and MICROTOPS and are shown in Table 5, all at a wavelength of 0.55 μm. As AERONET and MICROTOPS do not measure at 0.55 μm, logarithmic interpolation was carried out between 0.44 and 0.67 μm for optical depth and ω0.
Table 5. Comparison of Column Optical Properties at 0.55 μm From Aircraft and AERONET Sun-Photometer (Banizoumbou) for the Specific Profiles Shown in Figure 2a
AERONET1 and AERONET2 refer to Versions 1 and 2 of the inversion algorithm.
7.1. Determination of ω0
 In Table 4, data for mineral dust from each campaign are split into accumulation mode only (r < 1.5 μm) and accumulation plus coarse mode. The addition of the coarse mode to the Mie calculations decreases ω0 by 0.01. The most important observation is the high value of ω0 for “pure” mineral dust during DABEX and DODO-1, i.e., 0.99 with a variability of ±0.01. This value was consistent over multiple SLRs and several flights away from industrialized Niamey. Even close to Niamey the value of ω0 was 0.97 for dust below 1 km. SHADE showed a significant level of absorption, with measured ω0 as low as 0.95 on some days. It should be noted that the DABEX ω0 is based on the aerosol absorption properties at one midvisible wavelength only; since dust absorption has been demonstrated to increase rapidly at shorter wavelengths [Dubovik et al., 2002], it is not possible to label the DABEX dust “nonabsorbing”; we can only describe the optical parameters at 0.55 μm.
 In Table 5 the aircraft data are representative of the whole aerosol column (dust and BB aerosols) by the averaging of ω0 and α of the different layers, weighted to the optical depth contribution. There are relatively low values of ω0 for B161 compared to B160 and B165 because of the smaller dust contribution and hence dominance of BB aerosol. Such a difference is also seen in the AERONET data, although the two versions of the inversion algorithm produce significantly different values; that is, there is more absorption in Version 2. The agreement between Version 1 and the aircraft ω0 is good. The decrease in ω0 for Version 2 of AERONET relative to Version 1 was also indicated by Sinyuk et al.  for both BB and dust aerosol columns. But on the basis of the conditions encountered with the aircraft during DABEX, Version 2 produces erroneous levels of absorption. Note that since the AERONET 0.44 μm optical depth during B161 was 0.4 (i.e., less than the recommended value of 0.5 for accurate absorption retrieval), errors in ω0 will be greater than 0.03. No almucantar scans were available on the day of B164 (28 January 2006) because of partial cloud cover, hence no AERONET values of ω0 are reported.
 The distribution of ω0 at 0.55 μm in the vertical from all the SLR data in the vicinity of Niamey and Banizoumbou is shown in Figure 11. Broad vertical classifications of dust, BB and “mixed” aerosol have been added to Figure 11, with the “mixed” aerosol lying between 900–2300 m. The 900 m limit is the lowest height of all the aircraft profiles (flight B159 on 19 January) where we are confident (e.g., on the basis of measurements of α) that all aerosol below that limit is dust. The 2300 m limit is the greatest height from all the profiles of the lower limit of the BB plume (during flights B161 and B165); that is, we are confident that all aerosol above 2300 m is predominantly from BB, albeit mixed with a significant loading of dust. Within individual cases a distinct top to the dust layer can be seen in ω0, for example, in B165 at 2 km which coincides with Figure 3. The increase in the depth of the dust layer in B164 and B165 (28–30 January) coincides with an increase in dust optical depth (see section 7.3). Johnson et al. [2008a] show that the mean value of ω0 within aged BB aerosol was 0.86, and infer a value of 0.81 for BB that has had the dust component removed. These values, together with the “pure” dust from northeast of Niamey (0.99), are annotated in Figure 11. Within the dust layer, some points are below this 0.99 value owing to the influence of either (1) mixing with either BB aerosol; (2) industrial fossil-fuel sources in the Niamey conurbation; or (3) mineral dust exhibiting more absorption. There is also a decrease in ω0 with height in the dust layer which could be explained through interaction with BB aerosol, but in some individual cases (e.g., B163) this decrease is not evident. Likewise in the BB layer, there is sometimes a decrease in ω0 with height (B160) but sometimes not (B163). This variability could be attributed to the degree of mixing in the vertical, i.e., how uniform the aerosol properties become with time. Multiple BB layers from different sources have been identified during DABEX [Chazette et al., 2008] and this could explain some of the changes with altitude.
7.2. Determination of ke, g and α
 As seen in Table 4, the addition of the coarse mode in the DABEX Mie calculations decreased ke to less than half the submicron value owing to the greater particle mass. The effect on g, such as the increase in value by the addition of the coarse mode, can also be seen but the variation is relatively small. The values of ke and g were generally fairly similar between the three different campaigns.
Table 5 shows representative values of α from the aircraft nephelometer, Mie scattering calculations (using the measured size distributions as input), AERONET, and the hand-held MICROTOPS. The nephelometer values represent those calculated from scattering only; all the other methods represent values calculated for total extinction. Values of α were determined over the wavelength interval 0.45–0.70 μm, with the AERONET and MICROTOPS data being interpolated from their fixed wavelengths. As a comparison, values of α from SLRs during B160, B161 and B165 to the northeast of Niamey in heavy pure dust show variation between −0.20 and +0.04, i.e., lower than the column values in Table 5. The variation between the four flights is in agreement between all methods of determining α, i.e., smallest values during B165 (where dust contributed 75% of the optical depth) and greatest during B161 (where BB aerosol dominated the column near Niamey with the dust contribution at 35%).
 The aircraft values are consistently the lowest suggesting a bias to larger particles, and AERONET consistently the highest. The fact that the aircraft values show scattering rather than extinction cannot explain the differences because sensitivity tests using Mie calculations indicate that values of α reduce (by up to 20%) when absorption is included. Agreement between the aircraft and ground-based measurements becomes poorer as the mineral dust contribution to the optical depth increases. Geographical separation can account for differences, but the consistency in the differences in Table 5 hints at systematic reasons.
7.3. Aerosol Optical Depth
 For the three principal dust flights under scrutiny in this paper the influence of relative humidity on the aerosol optical depth through hygroscopic growth was small. Values of relative humidity were only significant (50–60%) close to the top of the BB layer and using the measured growth curves of Magi and Hobbs  within southern African regional haze, the contribution of aerosol water to the optical depth was calculated as typically a few %.
 AERONET optical depths are a fairly robust measurement with relatively high accuracy, compared to the column size distributions presented above which are the product of inversion and for which certain conditions are required for satisfactory inversions. Aircraft optical depths have been shown in the past to have variable levels of agreement with AERONET. Optical depths determined by integrating the nephelometer and PSAP in situ extinction have shown good agreement with AERONET as shown by Haywood et al. [2003c] for BB aerosol. Optical depths determined by integrating the PCASP Mie extinction, however, have led to poor agreement with the PCASP extinction consistently producing low values [Haywood et al., 2003c; Osborne et al., 2007; Johnson et al., 2008a]. These references describe confidence in the PCASP size distribution (e.g., agreement with AERONET volume distributions) but attribute the low Mie extinction to erroneously low total concentrations. Therefore it has been decided to omit PCASP optical depths from the present work.
Figure 12 shows a time series of aerosol optical depth at 0.55 μm as determined by various methods. The Banizoumbou AERONET daily means and daily ranges (interpolated between 0.44 and 0.67 μm) are shown together with the hand-held MICROTOPS daily means and ranges. There is good agreement between these two data sets in both the trends and absolute values. The aircraft optical depths (from profiles of nephelometer and PSAP) are shown in the solid diamonds, with the dust-only optical depth (i.e., from the low-level plume) shown by the open diamonds. The aircraft instrumental error bars represent typically an 8% error. The aircraft data have been adjusted to correct for (1) any atmosphere below the based of the profiles, and (2) large particle contribution that was not sampled by the nephelometer and PSAP. The latter was estimated by using the CVI PCASP-X size distribution above 1.5 μm. The aircraft optical depths range from 0.19 to 1.07, with the dust-only optical depth ranging from 0.07 to 0.81. The fractional contribution of the dust layer to the total aerosol optical depth varied between ∼6% and ∼80%, and this fractional contribution was 50% or greater on approximately half of all the aircraft profiles during the campaign.
 Despite some cases of good agreement (16, 19, and 23 January), the aircraft optical depths are biased to high values compared to AERONET, even taking in account the daily variability in AERONET. On some days, the dust-only optical depth is greater than the AERONET value (B161 and B165), suggesting a problem with either the aircraft or AERONET measurements which becomes more apparent when the dust fraction is high. As was found with comparisons of α, the agreement between between aircraft and AERONET optical depths is poorest where the dust contribution is greatest. Consistent with Figure 12, the individual profiles in Table 5 show that the optical depth was higher for the aircraft data compared to AERONET in all profiles during B160, B161, B164 and B165, although agreement was much better during B161 where the optical depth and the dust fractional contribution were relatively low. One reason could be that the large particle correction to the nephelometer data, up to 30% of the column optical depth, is incorrect and that the particle sampling was more efficient than has been estimated in previous studies [Haywood et al., 2003a]. Not applying the large particle correction brings the aircraft optical depths in Figure 12 in much closer agreement with the AERONET data.
 This study has summarized aircraft in situ measurements made within mineral dust plumes in Sahelian west Africa in January 2006. These dust plumes were always below 2 km and often shallower than this; that is, they occupied the boundary layer which was also observed thermodynamically by a temperature inversion and increase in dew point. Although the overlying aerosol layer of aged biomass burning aerosol contained externally mixed dust particles, the concentrations were at least 1 order of magnitude lower than the main dust layers and so we have ignored their optical contribution in this paper.
 Confidence in labeling certain conditions as “pure” mineral dust derives from the: (1) shape of the aerosol number and volume size distribution across the accumulation and coarse modes; (2) PSAP aerosol absorption and nephelometer scattering, with a single scattering albedo close to unity; (3) nephelometer Angström exponent, α, which was often close to zero; (4) modeled air mass origin analysis at low-level showing the flow of the Harmattan wind toward the south; (5) bulk and single particle filter analysis [Chou et al., 2008; P. Formenti et al., Composition of mineral dust from Western Africa: Results from the AMMA SOP0/DABEX field campaign in Niger, January, 2006, submitted to Journal of Geophysical Research, 2008]; and (6) AMS chemistry [Capes et al., 2008] showing very low mixing ratios of organics and sulphate.
 The relationship of CO and O3 mixing ratios from flights B160 and B161 showed a strong positive correlation within the BB aerosol, but this relationship broke down within the low-level dust layer. This loss could be due to either deposition to the ground, destruction on the surface of airborne mineral dust, or long-range transport from polluted regions. The significant concentrations of CO within the dust layers suggest a history of combustion from BB or industrial sources, but with no evidence of aerosol absorption from black carbon then this influence was sufficiently old enough to allow removal of all combustion aerosol but not the CO.
 The accumulation and coarse modes were clearly identified from volume distributions of the dust but the number distributions were much flatter. Two log-normal curves were used to adequately described the dust aerosol between 0.05 and 5 μm radius, both for aerosol number and volume. Although greater accuracy was achieved with a five log-normal fit, two modes are potentially more useful when computational costs are at a premium [e.g., Greed et al., 2008]. It is worth noting that whilst other dust studies usually have their smallest dust mode at 1 μm, our smallest mode at 0.075 μm suggests the possibility that at least a fraction of the particles are not mineral dust but of some other composition. Even if this is true, we know that the composition does not contain black carbon on the basis of our absorption measurements.
 Comparison of the volume distributions representative of the aerosol column with those retrieved from AERONET show some level of agreement, particularly for Version 2 of the inversion algorithm. There were only minor discrepancies in the position and width of the coarse mode, but significant discrepancies in the balance between accumulation and coarse modes were evident. The solution to be AERONET inversions maybe not truly representative of the aerosol column size distribution because of the effect of having a BB layer (dominated by submicron partially absorbing particles) overlying a dust layer (dominated by supermicron nonabsorbing nonspherical particles). In terms of the vertical distribution of the aerosol, it should be noted, Dubovik and King  state that this will only have a very small effect on the retrieved size distributions. Another source of error lies in calculating the aircraft size distributions that were taken from SLR means within the significant aerosol layers which were then averaged together on the basis of the optical contribution of each layer. This assumes that the SLR means, which were fixed at specific altitudes, are representative of the aerosol layers.
 A mean single scattering albedo of 0.99 ± 0.01 within the dust plumes away from industrial regions shows that there was little or no absorbing material (relative to midvisible light) naturally occurring within the dust. It was hypothesized before DABEX commenced that internal mixing between BB aerosol and mineral dust particles would likely occur and so lead to complex optical properties. Within the elevated BB layers where aged dust (exhibiting a relatively small coarse mode) and BB particles were mixed together, the filter analysis did not reveal any evidence of internal mixing [Chou et al., 2008]. Therefore our assumption, like the assumption used by Johnson et al. [2008a], of external mixing is justified. The decrease in ω0 with height within the bottom 2 km of the atmosphere is indicative of BB aerosol mixing down, or being entrained, into the dust layer. The Saharan dust observed by Haywood et al. [2003a] off the coast of Senegal during SHADE in September 2000 had a mean ω0 of 0.97 with a range of 0.95–0.99. With a DODO-1 mean of 0.99, which took place after DABEX also off the coast of Senegal, it shows that the absorption varies with the specific source region of the dust. Very high values of ω0 for mineral dust are not unusual: for example, Clarke et al.  found values for submicron dust of 0.99 within Asian outflow over the sea during ACE-Asia, and more relevant, Todd et al.  retrieved values of 0.98 from AERONET for dust from the Bodele depression.
 The coarse modes during DABEX were not particularly large, even though the low-level plumes were probably quite young, which is most likely related to wind speed. Although the addition of the coarse mode in calculations of ω0 and g has only a small effect, McConnell et al.  show a dust plume example during DODO-2 (September 2006) where a strong coarse mode had a large influence on ω0, i.e., a reduction from 0.98 for accumulation mode only to 0.90 for inclusion of the coarse mode. This DODO-2 case study displayed a much larger coarse mode than those from either DODO-1 or DABEX. The variability of ω0 for dust in the literature from both aircraft and AERONET data sets can be reconciled as much by variations in the large particle content (whether real or an artefact of sampling) as by variations in the absorption properties of the dust particles themselves. The specific extinction (ke) during DABEX was heavily influenced by the mass contained in the coarse mode, i.e., ke halves in value, and the optical depth as we have seen is also significantly affected by coarse mode dust.
 Good agreement was found between the AERONET version 1 ω0 and the column-weighted aircraft data for specific profiles over a number of flights; version 2 produced consistently too much absorption. In terms of aerosol optical depth, despite some good agreement on occasion, the aircraft vertical integrations of extinction coefficient produced values that were consistently higher than AERONET and the hand-held MICROTOPS. Although there are undoubtably errors due to geographical separation of the Banizoumbou site and the aircraft profiles (which are heavily slanted), it is hard to reconcile a consistent bias in the data. Our corrections for the large particle contribution are perhaps too great; that is, the large particle sampling by the nephelometer could be more efficient than previous studies have suggested. Implicit in our corrections are the assumptions that the CVI inlet (for measuring the large particle size distribution) samples efficiently and the nephelometer sample inlet (a Rosemount) is very inefficient. Making the assumption that the nephelometer efficiently samples particles up to 3–4 μm radius can account for a large part of the discrepancy between the aircraft and AERONET optical depths.
 Considering the difficulty there is in measuring large particles with an aircraft, and the unknown sampling errors in the present work that lead to errors in the optical properties, this is an area that would require serious attention in future dust studies. Recently installed on the BAe-146 is an externally mounted and noninstrusive Small Ice Detector (Mark 2) that, as well as quantifying the nonsphericity of ice crystals down to 1 μm, should also be able to perform similarly with aerosol particles. Flight testing of this instrument is now under way for its use in future aerosol campaigns.
 The AMS data were kindly provided by Gerard Capes (University of Manchester). FAAM is jointly funded by the Natural Environment Research Council and the Met Office. The authors would like to thank the staff at FAAM, DirectFlight, and Avalon during the aircraft operations for their enthusiasm.