During August 2015, the UK BAe 146-301 Research Aircraft operated by the Facility for Airborne Atmospheric Measurements was stationed at Praia on the Cape Verde islands, for the ICE-D and AER-D field campaigns. The AER-D project comprised six flights, focusing specifically on dust properties within and beneath the SAL. Each flight was performed 4 to 5 days apart, due to the transported dust in the SAL displaying intermittent character, typical for summertime dust export (Jones et al., 2003; Schepanski et al., 2017). Details of the flights are shown in Table 2, and flight tracks are shown in Fig. 1.

The events sampled revealed accumulation mode AODs at 550 nm (see Sect. 2.3.1) from 0.4 to 0.8, which is within the range expected from August climatology over the eastern tropical Atlantic indicated by satellite imagery (not shown). The aircraft observations revealed a mostly typical vertical dust structure (see Sect. 3.1). The exceptions were b923 and b924, a pair of flights which observed an intense dust outbreak with AODs of up to 2.5 at around 24∘ N off the coast of West Africa, with in situ and remote-sensing observations showing a different and complex vertical structure compared to the conventional SAL model of an elevated dust layer (discussed in Marenco et al., 2018).

Two flights (b928, b934) formed part of the Sunphotometer Airborne Validation Experiment in AER-D (SAVEX-D) project, focused on providing airborne measurements for comparison with two types of ground-based sun photometer. Requirements for non-cloudy skies and moderate to high dust loadings and proximity to the ground-based sun photometers required that the first SAVEX-D flight (b928) took place close to Praia, while the second (b934) took place close to the island of Sal. Three flights (b923, b924 and b932) were aimed mainly at mapping the vertical and horizontal aerosol structure. These flights headed to the northeast in order to encounter heavy dust loads closer to the African coast on 12 August. Flight b920 was conducted near the Cape Verde archipelago and co-located with the path of the International Space Station (ISS), in order to fully characterize the SAL and validate the Cloud-Aerosol Transport System (CATS) remote-sensing lidar instrument onboard the ISS.

Each flight consists of a combination of sloped profiles (abbreviated to “P”, e.g. “P1”) and straight-and-level runs (known as “SLRs” and abbreviated to “R”, e.g. “R2”) at various altitudes, generally selected to be within the aerosol layer of interest or at altitudes appropriate for radiometric measurements. Here we present results from 31 profiles and 19 in situ aerosol sampling SLRs, as shown in Fig. 1. Five SLRs were performed in the marine boundary layer (MBL) at 30 to 35 m above sea level (one per flight), and 14 SLRs sampled the SAL at altitudes between 1.8 and 4.1 km. Exceptions were an intermediate layer sampled during flight b920 containing dust at 1.2 km and heavy dust sampled during flight b924 at 920 m. Full information about profile and SLR times and altitude are available from the Centre for Environmental Data Analysis (see the “data availability” section).

Broad geographic dust source locations, where dust was uplifted before being sampled by the aircraft during AER-D, have been identified using Spinning Enhanced Visible and Infrared Imager (SEVIRI) “dust RGB” (red–green–blue) thermal infrared satellite imagery product (Lensky and Rosenfeld, 2008), where dust events are identifiable by their bright pink colour (e.g. Brindley et al., 2012). Dust events sampled by the aircraft are tracked backwards in time visually until uplift times and locations are identified, identical to Ryder et al. (2013b). The high temporal resolution (15 min) of the imagery enables easier tracking compared to other spaceborne sensors which may have only two overpasses per day. However, this method is subjective, and therefore we allow generous errors in terms of geographic dust uplift location, and dust uplift time. Additionally, no height resolution information is available from the imagery, so source areas are categorized for each flight and not for each sampling altitude within flights.

The examination the dust events sampled during AER-D reveals that the dust in every event was initially uplifted by a mesoscale convective system (MCS) and a resultant cold pool (haboob) which spread out radially. This is in keeping with recent findings that cold pools are the dominant mechanism for summertime Saharan dust uplift (Marsham et al., 2013; Allen et al., 2013). The dust events then took 1 to 5 days to reach the Atlantic Ocean where they were sampled by the aircraft. In three cases, additional uplift events could be observed merging with the initial dust event, and in these cases error bounds on dust age were set to include all possible uplift events and were therefore very broad. In most cases, convection and clouds could be observed developing over the SAL during the afternoon in the top of the dust layers during their transport. Therefore, it is possible that the dust observed has been processed by clouds (e.g. Ryder et al., 2015 Sect. 4.1.4; Diaz-Hernandez and Sanchez-Navas, 2016), though no clouds were present when sampling took place.

The SEVIRI imagery is not able to give altitude-resolved information and can be subjective, particularly when dust loadings are light, at low altitude or in a moist environment, making dust appear less pink and more difficult to identify (Brindley et al., 2012). This is more evident in the dust tracked for flights b932 and b934 where dust loadings were lower. This introduces a small level of uncertainty into both the source locations and dust ages, which we account for by giving generous error bars to the dust uplift times and source locations. Hybrid Single-Particle Lagrangian Integrated Trajectory model (HYSPLIT) back trajectories (Draxler and Hess, 1998; Stein et al., 2015) were also run for the AER-D dust events. In only one of the five dust events was the dust source location similar to that observed in the SEVIRI imagery. In every case the back trajectories indicated a transport path and transport time different to that shown by the SEVIRI imagery. Although the SEVIRI methodology has its limitations, the back-trajectory method results were clearly not compatible with the information from SEVIRI. Therefore, back trajectories are not used to determine source location or age here. Additionally, another limitation of back trajectories is that they only indicate when an air mass nears the surface but do not reflect potential uplift conditions (e.g. surface wind strength or soil conditions). It has been shown that models and reanalyses are currently unable to adequately represent convective events and winds over the Sahara and Sahel, particularly due to the challenges of representing cold pools. For example, Garcia-Carreras et al. (2013) examine the role of convective cold pools and suggest that “the misrepresentation of moist convective processes can affect continental-scale biases, altering the West African monsoon circulation.” Many other publications have examined the misrepresentation of Saharan convective events (Marsham et al., 2011; Heinold et al., 2009; Sodemann et al., 2015; Trzeciak et al., 2017; Allen et al., 2015; Roberts et al., 2017; Engelstaedter et al., 2015). Since convective events are the drivers of dust uplift in all the AER-D cases, we do not consider HYSPLIT back trajectories (with relatively low model resolution of half a degree) to be informative here due to the challenges the models face in representing Saharan circulation. Finally, we note that back trajectories are recommended to be used with caution for dust events over the summertime Sahara (Trzeciak et al., 2017).

Much of the instrumentation operated during ICE-D was identical to that operated during the Fennec campaign, described in Ryder et al. (2013b, 2015). Relevant information is provided below, noting where instruments and processing differed. Throughout this article all particle sizes are referred to in terms of diameter (d), and optical properties are presented at 550 nm unless stated otherwise.

Filter samples were analysed by scanning electron microscopy (SEM, instrument model JEOL JSM 6301F) coupled to an X-ray energy-dispersive spectrometer (Silicon Drift X-Max 80 mm2 Detector and Aztec Advanced-INCA350 analyzer, Oxford Instruments) to provide information on the coarse and fine fractions respectively. Analysis was performed on a portion of filter cut and mounted on an aluminum stub using a double-sided adhesive and then covered with a thin film of platinum (Pt) by sputter coating (Jeol JFC 1100E). Particles were found to be evenly distributed across the filter sample. Images were acquired by a series of transects at two magnifications (×2000 (55.9 nm pixel-1) and ×10000 (11.0 nm pixel-1)), scanning between 397 and 1116 particles per sample.

Particles were sized by processing the SEM images, which are essentially 2-D projections of 3-D particles. The 2-D projections were fitted with a circumscribed ellipse, to produce an ellipse area-equivalent diameter, where the major and minor axes (dmax and dmin) are used to define the particle diameter d according to d=(dmaxdmin)1/2. The aspect ratio was accordingly calculated using dmax/dmin.

It was not possible to use automated image contrast to calculate projected particle area because of a high degree of variability in particle contrast. Our filters sizing this technique may oversize the particle size for two reasons. Firstly, the area of a fitted ellipse may be larger than a projected particle area, though the particles were not noticeably jagged around their edges. Secondly, our method may oversize particle volume where the shape is a platy silicate and has a tendency to fall with its largest surface parallel to the substrate. For example, Chou et al. (2008) found the height of dust particles examined under SEM to be around one-third of their major axis length. Additionally, we tested the sensitivity of the filters PSD to using the mean XY and CC sizing methods applied to the OAP data (not shown). Using a mean XY method on the filter data did not produce significantly different results, while using the CC method was found to shift the PSD towards larger particles, similar to the findings from the OAP size metric comparisons.

Semi-quantitative elemental chemical analyses integrate the following elements: Na, Mg, Al, Si, P, S, Cl, K, Ca, Ti, Mn and Fe. The results are expressed as a weight percentage of the associated oxides (Na2O, MgO, Al2O3, SiO2, P2O5, SO3, Cl, K2O, CaO, TiO2, MnO, Fe2O3) normalized to 100 %. The composition type of each particle is then identified based on these percentages as belonging to one of the following categories: alumino-silicates, quartz, sulfate, salt, thenardite, gypsum, Ca-rich, Ti-rich, Fe-rich and others. These are interpretations based on the elementary chemical analysis. For example, if we observe the simultaneous presence of calcium and sulfur (typically within 10 % of one another), this could be several possible minerals: anhydrite (CaSO4), bassanite (CaSO4 ⋅ 0.5(H2O)) and gypsum (CaSO4, 2H2O), though the most probable atmospheric form is gypsum. Sulfate is classified by the dominance of SO3. Thenardite is classified when both Na2O and SO3 are dominant. Salt is classified by percentages of both Na2O and Cl being high, indicating halite particles of marine origin. Alumino-silicates, quartz, thenardite, gypsum and Fe-rich particles indicate terrigenous particles in these cases. Only one black carbon chain-like structure was observed during the analysis of over 6500 particles, and therefore this aerosol category is not included.

Although particles may frequently be a mix of several composition types, they were classified according to their component type which made up the greatest oxide percentage. This may be a particularly important assumption regarding the iron component, which was consistently present in small amounts on most particle types but only classified as iron when iron was the dominant component; i.e. in general when iron oxide constituted over 50 % of the composition. More complex methods of accounting for iron as a minor component are possible (e.g. Kandler et al., 2011a; Balkanski et al., 2007) but are beyond the scope of this work. Although composition information is available for all particles sampled under magnifications of ×2000, many particles sampled under the high magnification did not provide a strong enough signal to provide composition data. Thus, the number of particles available for composition analysis in the fine fraction is much lower than that for the coarse fraction.

Filter sample analyses use subsets of data from each magnification in order to take advantage of the best counting statistics for each size range and optimal viewing at each magnification. Filter sample size distributions are calculated using fairly finely resolved size bins from 0.05 to 40 µm as shown in Fig. 7. However, at each magnification, sensitivity to the smallest particles detected is low due to a low SEM signal (fewer photons emitted for smaller volume particles). Therefore, particles smaller than 0.1 and 0.5 µm are excluded for ×2000 and ×10000 respectively. Additionally, for the ×10000 magnification, not many particles larger than 1 µm are counted in comparison to ×2000, so particles larger than 1 µm are also excluded.

For aspect ratio, composition and refractive index analyses, particles sized 0.1 to 0.5 µm are taken from the ×10000 magnification, while particles sized between 0.5 and 40 µm are taken from the ×2000 magnification and scaled appropriately to account for the different substrate areas examined under the different magnifications. Size-resolved composition data utilizes six size bins with edges at 0.1, 0.5, 1.0, 2.5, 5, 10 and 40 µm. Information for the full filters size distribution accordingly covers 0.1 to 40 µm. Additionally, bulk filter sample properties are calculated specifically for the accumulation mode, covering diameters 0.1 to 2.5 µm, in order to replicate the in-cabin measurements behind the Rosemount inlets and also for the bulk sample from 0.1 to 40 µm to cover the full size distribution. Aspect ratios are presented as number fractions; composition is presented as volume fraction.

Optical properties (SSA, MEC and g) are calculated using a Mie scattering code, where dust particles are assumed to be spheres. Size distributions used are taken from the wing probes, representing either the FULL PSD or the ACC PSD. Refractive indices are taken from either the iterated RI method (Sect. 2.5.1) or from internal or external mixing RI values calculated from the filter samples (Sect. 2.5.2).

When the iterated RI values are used, composition, and therefore k550, is assumed to remain constant with particle size. This allows the iterated RI for the accumulation mode to be applied to the full size distribution. For all cases m550 was assumed to be 1.53. Iterated values of k550 specific to each SLR were used. For the MBL k550 varied between 0.0001 and 0.0005, and for the SAL k550 varied between 0.0005 and 0.0025 (as shown in Fig. 3). Although a value of 1.53 for m550 is higher than that produced by the filter sample composition results (Sect. 3.5, 1.47–1.49) the filter result is likely biased low due to the reasons discussed in Sect. 3.5. We also performed a sensitivity test to using m550 of 1.48 and 1.58 and found that deff changed by up to 5 % and SSA by under 1 %.

Where composition from filter samples is used, RI is allowed to vary as a function of particle size according to the composition results. For internal mixing assumptions, the refractive indices calculated according to Sect. 2.5.2 are used. For external mixing assumptions, size distributions of each mineral component are calculated by weighting the size distribution from the wing probes for each mineral component using its number fraction. Scattering properties for the size distribution of each mineral are then computed using Mie scattering code and the same literature refractive indices as described above. Summing scattering and absorption over all minerals then provides the total optical properties for the external mixing case.

Although it is clear that dust particles are not spherical, sensitivity of SSA to shape was tested by Otto et al. (2009) and Johnson and Osborne (2011), who found that SSA changed by under 1 % and 2 % respectively when non-spherical particles were assumed. This is less than our uncertainty in SSA of 5 % given above due to the refractive index, and therefore we consider this an acceptable assumption.

Figure 1 shows the dust uplift locations (sources) determined using SEVIRI RGB imagery for dust sampled during each AER-D flight. Table 4 shows the corresponding meteorological event driving the uplift and also the age of dust when sampled by the aircraft. Dust sampled by the first two flights, b920 and b923/924 originated from large-scale dust events uplifted in easily identifiable single events. The b920 dust was uplifted over southern Algeria (blue circle in Fig. 1) and subsequently transported northwards and then westwards. The dust emerged over the Atlantic from southern Morocco, being sampled by the aircraft after around 4 days' transport. The b923/b924 dust was uplifted further west, over northern Mali (green oval in Fig. 1), and was also subsequently transported northwestwards, emerging over the Atlantic over southern Morocco and sampled by the aircraft after a shorter time of 2 days due to the more direct transport path.

During the days leading up to flight b928, there was a change in the dominant meteorological dust export mechanism, as described by Liu et al. (2018). First, dust was uplifted over northern Mauritania (orange circle number 1) underneath widespread altocumulus cloud, which moved slowly to the southwest. Here, it became mixed with dust which had originated from northern Mali (orange circle number 2) where it had been uplifted by an MCS and haboob. This moved rapidly westwards, driven by strengthened 700 mb winds, which characterized the second phase of ICE-D (Liu et al., 2018). The two dust events became mixed together and subsequently transported dust over the Atlantic via the Mauritanian coastline, thus taking a more direct southerly transport route than the dust exported and sampled by the first two flights of the campaign (b920 and b924). Thus, the dust sampled during b928 was a mixture of dust from two different source regions. The final two flights, b932 and b934, consist of dust uplift events which subsequently overpassed secondary (or tertiary) uplift events and are therefore considered as dust from a mixture of the identified sources. The dust sampled during b932 was initially uplifted by an MCS and haboob close to the Mali–Algerian border, and some small-scale reinvigorated convection caused additional uplift over northern Mali before the dust was transported southwestwards towards the Mauritanian coastline. Dust encountered during b934 was also initially uplifted in almost the same region along the Mali–Algeria border by an MCS and haboob, and on each subsequent evening a new convective cell developed over the transported dust with a new haboob: over the Mali–Mauritanian border (purple circle number 2) and then over western Mauritania (purple circle number 3). Therefore, the range of possible dust ages for b934 is very large (16 h to 3 days). The range of sources identified during AER-D is consistent with well-known source regions in the literature (Engelstaedter et al., 2006; Formenti et al., 2011; Evan et al., 2016; Scheuvens et al., 2013), particularly the Mali–Algeria border hotspot downwind of the gap between the Atlas and Ahaggar Mountains (Potential Source Area 3 in Formenti et al., 2011), which contributed to four of the five AER-D dust events.

Figure 4 shows the vertical structure of the five dust events sampled during AER-D from in situ aircraft measurements. Four events (b920, b928, b932 and b934) display typical SAL structure, with elevated dust from a base of between 0.5 and 1.5 km up to an upper bound of 5 km (6 km in one case) overlying the MBL. Accumulation mode extinction coefficients vary from around 100 to 500 Mm-1 in this layer, sometimes with fairly constant values in the vertical (b920, b934). Other layers were more sinuous (b932, b928) but always had accumulation mode scattering Ångström exponents (SAE; 700 to 450 nm) of a fairly constant value around -0.5 – a clear indication of coarser particles. Within the SAL, water vapour mixing ratios were low (under 10 g kg-1) and potential temperature increased slowly with altitude. Relative humidities were low at 30 %–50 % at altitudes where SLRs were performed. Contrastingly, in the MBL relative humidities were high (> 90 %) as expected, with the MBL capped by a temperature inversion. SAE values in the MBL were variable, sometimes exhibiting a jump to positive values (b920, b934) suggesting a dominance of smaller particles, whereas sometimes hardly displaying any difference to the overlying dust (b928, b932). In the former two cases, the PCASP size distribution confirms a greater relative contribution from fine (d < 0.3 µm) particles in comparison to those sized 0.3 to 3 µm. This vertical structure is as expected for the region in summer and in keeping with the elevated and dusty SAL above the MBL (Prospero and Carlson, 1972).

In contrast, the dust event sampled on flights b923 and b924 displays very large dust loadings and different vertical structure. The upper SAL, from 2.5 to 5 km, displays a roughly constant potential temperature and water vapour mixing ratio, with extinction coefficients of around 140 Mm-1. Between 0.5 and 2.5 km, dust loadings become extremely high, reaching 2122 Mm-1 at 1.2 km and contributing to the very high AOD of 2.54 for the accumulation mode only (including coarse particles would increase the estimated AOD further). To the authors' knowledge, this is the highest aircraft-measured value of dust-related extinction measured for dust transported over the ocean, and it is explored further by Marenco et al. (2018). (Note that lidar-derived AODs and extinction shown by (Marenco et al., 2018) are slightly lower than those shown here, which may be due to different extinction properties of the dust at the lidar wavelength of 355 nm, the Rosemount inlet enhancement effects shown in Fig. S2, or the differences between a lidar curtain and sloped aircraft in situ profile.) Moisture levels here decrease with decreasing altitude (8–12 g kg-1) and are larger than observed within dust during the other flights. The higher levels of moisture are perhaps not surprising given that an MCS haboob uplifted this dust 2 days earlier, and 1 day earlier Saharan boundary layer (SABL) convection can be seen impacting the dust over northern Mauritania (consistent with Marsham et al., 2013; and recycling of moisture within the dusty SABL in Ryder et al., 2015). Note that these values of extinction do not include the coarse mode, and therefore actual extinction values will be even higher. Beneath the thick, low-altitude dust layer was a shallow MBL extending up to 500 m. This unusual vertical structure of intense, thick dust within the bottom half of the SAL is not in keeping with the conventional SAL model (Prospero and Carlson, 1972) of well-mixed elevated dust throughout the whole SAL and is therefore of additional interest. This is discussed in detail by Marenco et al. (2018).

Figure 5 shows the wing probe size distributions for each SLR, separated into those measured in the SAL and MBL. The size distribution in the SAL displays a broad shape which does not change with increasing or decreasing dust load but simply shifts between higher or lower volume concentrations. For example, for b924 under large dust loadings where lidar-derived AODs approached 2.0 (Marenco et al., 2018), the volume concentrations are markedly larger than on other flights in AER-D (green points in Fig. 5), although the size distribution shape is much the same. This is in contrast with measurements over land close to dust sources during Fennec (Ryder et al., 2013b) where the absence or presence of the coarse and giant modes had a strong impact on the overall shape of the size distribution, since the relative proportion of giant particles was observed to increase. The peak of the volume concentration during AER-D was constantly between 5 and 10 µm diameter. We observe a fine mode of aerosol at 0.1–0.3 µm, which is evident when dust concentrations are lower but becomes eclipsed by the accumulation mode dust during flights with larger dust loadings (e.g. b924). Section 3.4 shows that the composition of d < 0.5 µm aerosol in the SAL during AER-D was dominated by sulfates and salts, thus explaining the different behaviour of this mode. Liu et al. (2018) examine the composition and behaviour of the accumulation mode during ICE-D in more detail.

In the MBL, a distinct giant mode is evident between 20 and 60 µm during three flights (b924, b928 and b932: red, green and orange in Fig. 5b). For flight b928, filter sample analysis (Sect. 3.4) confirms that this giant mode is composed of dust rather than sea salt. Higher wind speeds in the MBL, which may be an indicator of sea salt abundance, are not correlated with the presence of this giant mode (10–11 ms-1 for b924 and b932; 3–5 ms-1 for b920, b928 and b934). It is possible that turbulent mixing from the SAL to the MBL is more likely given a weak inversion and higher turbulence. Therefore, we also examined whether the strength of the temperature inversion (calculated using the vertical gradient of potential temperature) and the strength of turbulence at the inversion (indicated by the variance of vertical velocity) could be related to the giant particles present in the MBL. However, we found no obvious connection in these cases. We note that giant mode particles in the MBL were only present during flights when there was a significant presence of giant mode particles in the SAL above: there is a giant mode in Fig. 5b (flights b924, b928, b932; green, orange, red) only when a higher concentration of particles in the SAL in the same size range was measured (Fig. 5a, same colours). When few particles are measured in the SAL in this size range (b920 and b934, blue and purple), the MBL giant mode is also absent. Thus, the observations suggest that the giant MBL mode may be dust being deposited from the overlying SAL towards the ocean. This has also been suggested by Jaenicke and Schutz (1978) from aerosol surface observations at Sal, Cape Verde, where giant particles (d > 40 µm) were observed to arrive at the site a day after coarse dust particles (6 < d < 60 µm).

Despite careful error analysis and selection of instrumental data, there is still a large degree of noise as a function of diameter in the wing probe size distributions shown in Fig. 5, which was also not indicated in the filter sample PSDs. Therefore, in Fig. 6, lognormal curves using four lognormal modes are fitted to the wing probe instrumental data using a least squares regression (Markwardt, 2008) with mode parameters from Table 5. The lognormal curves represent a best fit across the full size range for the instruments available. Differences between the effective diameter calculated with the best-fit lognormal curves and the observed PSDs are between 10 % and 15 %, resulting largely from deviations between the observations and best-fit curve in the CDP size range (3 to 20 µm).

It can be seen that the volume concentrations in the SAL are larger than those in the MBL, as expected due to the higher concentrations of elevated dust and that the SAL size distribution has a notably different structure. Dust in the SAL displays a broad size distribution with contributions from particles over a wide range of sizes (0.3 to 100 µm), peaking at 5–10 µm. The MBL size distribution has a narrow peak at d∼5 µm and a giant mode contribution at around 20 to 60 µm, where volume concentrations in the MBL mean are actually larger than the SAL mean. Compared to PSDs over the desert measured with aircraft during Fennec and SAMUM1 (Ryder et al., 2013b; Weinzierl et al., 2009), where the volume distributions peak at diameters larger than 10 µm, AER-D has a smaller giant mode contribution, as would be expected. The AER-D PSD is more in keeping with other aircraft observations of transported dust, where the volume distributions peak at diameters between 3 and 10 µm (measured during the following: the Geostationary Earth Radiation Budget Intercomparison of Longwave and Shortwave radiation (GERBILS) – Johnson and Osborne, 2011; Aerosol Direct Radiative Impact on the regional climate in the MEDiterranean region (ADRIMED) – Denjean et al., 2016; SAMUM2 – Weinzierl et al., 2011; the Saharan Aerosol Long-range Transport and Aerosol-Cloud-Interaction Experiment (SALTRACE) – Weinzierl et al., 2017). Despite some agreement here, the variation in size across the 3 to 10 µm size range can still have a large impact on the dust radiative effect (Tegen and Lacis, 1996). MBL PSDs under the influence of dust advection were measured by Kandler et al. (2011b) at Cape Verde. Their observations revealed a sharp mode at 10 µm, declining steeply at larger sizes, which contrasts with our modes centred at d∼5 µm and 20 to 60 µm. They also found that when air masses had a maritime origin (but still dominated by dust), the PSD was broadly flat between around 10 and 80 µm, similar to the AER-D MBL PSDs.

Size distributions have also been derived from the four filter samples and are contrasted with the wing probe size distributions in Fig. 7. Such comparisons have previously been shown to be challenging due to the different nature of measurement from each instrument (e.g. Chou et al., 2008; McConnell et al., 2008; Price et al., 2018), and discrepancies are common, particularly with non-spherical particle geometry. We note that particles measured by the PCASP will be randomly orientated due to passing into the PCASP nozzle. Larger particles sampled by the CDP and those measured by the 2-DS and CIP15 may be aligned horizontally in the atmosphere (e.g. Ulanowski et al., 2007) and measured in this orientation. Also, each technique allocates size using a different methodology. Additionally, filter sample viewing is likely to preferentially view the larger cross section of plate-like particles as they fall flat on the filter substrate.

Figure 7 shows that in most cases, the filters size distribution is greater than that from the wing probes. In the size range 0.5–1 µm the best agreement is found. In three of the four SLRs, the filter coarse-mode volume distribution exceeds that of the wing probes by an order of magnitude or more. At d < 0.5 µm the filter size distribution shows a much more distinct fine mode than that seen from the PCASP. Solid lines for the filter samples (blue and orange) in Fig. 7 indicate volume distribution calculated assuming a spherical shape using diameter calculated from the area-equivalent diameter of the fitted ellipse. It is possible that the filter samples overestimate size, and therefore also volume, if particles are plate-like and fall flat on the filter substrate. We test whether accounting for this by including a representation of particle height, using a height : maximum axis ratio of 1:3 (Chou et al., 2008), can improve the agreement with the wing probes. This sizing metric is shown by the dashed lines in Fig. 7. Although accounting for preferential particle orientation on the filters makes some differences to the derived size distribution, it is not sufficient to allow agreement with the wing probe size distributions. It is possible that there were problems with the filters flow rates measured during AER-D resulting from some combinations of filter pore sizes and filter supports. SLRs with higher flow rates (b920 R2 and R5) show better agreement with the wing probes (Fig. 7a and c) and also used filter samples with the larger pore sizes (0.4 µm), compared to the other two SLRs (Fig. 7b and d) which have worse agreement. We note that for all the SLRs shown, the shape of the coarse-mode size distribution is the same for both filters and wing probes even if they are offset. For example, the broad shoulder of the coarse-mode size distribution can be seen in b928 R2 (Fig. 7b), and the sharper drop off of the coarse mode can be seen in b920 R5 and b934 R6 (Fig. 7c and d) for both filters and wing probe size distributions. Additionally, diameters of the peak size distribution (5–10 µm) are consistent in filters and wing probes for the SAL (c and d). We also note that the filters PSD is much smoother than that from the CDP.

The examination the metric of effective diameter is useful, since it takes into account the contribution of a range of particle sizes. Figure 8a shows that the effective diameter of the full size distribution computed from the wing probes is fairly constant with altitude with no discernible trend, as expected for a well-mixed SAL. A similar picture is seen for the VMD but with values of around 5 to 6 µm (not shown). No evidence is found for larger particles being more abundant closer to the base of the SAL, as would be expected due to gravitational settling. For the SAL the mean (minimum, maximum) deff value is 4.0 µm (3.6, 4.7) while for the MBL the mean is 4.6 µm (3.4, 5.5). Variation within the MBL is much greater due to the absence or presence of the giant mode shown in Fig. 6. Mean (minimum, maximum) deff for the accumulation mode is smaller at 1.7 µm (1.4, 2.0) for the SAL and 1.1 µm (0.7, 1.4) for the MBL, in agreement with Liu et al. (2018), reflecting the enhanced fine mode in the MBL size distribution from 0.1 to 0.3 µm diameters shown in Fig. 5. We also note that deff for flight b924 in the thick dust layer was 4.3 µm – not at all different from other deff values in typical dust loadings, despite total volume concentration being much larger. Mean AER-D size parameters of deff and VMD are given for the SAL and MBL in Table 6.

In addition to deff, we also show the maximum size (dmax) detected by the 2-DS XY at concentrations greater than 10-5 cm-3 (see Sect. 2.3.2) as a useful indication of transport of the largest sizes, which can contribute substantially to the mass fraction and are therefore important to dust biogeochemical cycles. Figure 8b shows the maximum size detected during AER-D. The largest value of 80 µm at 900 m comes from flight b924, during the intense dust event. Within the SAL, dmax varied between 20 and 80 µm, and the same range was found within the MBL. There is no clear trend of dmax decreasing with altitude. Particles sized 20 µm (40 µm) or larger were detected in 100 % (36 %) of the AER-D dust layers investigated and in 100 % (80 %) of MBL layers. The lack of decrease in dmax with altitude during AER-D is similar to that observed over the desert during SAMUM1 by Weinzierl et al. (2011).

The prevalence of coarse particles shown in Fig. 8b is greater than predicted due to settling velocities alone: a 20 µm particle should fall 5 km in 1.4 days (Li and Osada, 2007), and a 40 µm particle would take 13 h for the same distance. Therefore, with the dust age range in AER-D estimated at 17 h to 4.6 days, we would not expect any 40 µm particles to be present at all and would only expect 20 µm particles at altitudes below 2.4 km, yet 40 µm sizes were measured over 3 km and 20 µm sizes were measured at altitudes over 4 km.

The aspect ratio derived from the filter samples, defined as the ratio of the major to minor fitted-ellipse axes, is shown in Fig. 9. The results are somewhat noisy in the larger size ranges due to the relatively small number of particles analysed, and samples where n < 10 have been excluded. As expected, larger particles are more non-spherical in general, with higher median aspect ratios (between 1.30 and 1.51) for the 5 to 10 µm and 10 to 40 µm size ranges. This contrasts with the 0.5 to 5 µm range where median aspect ratio varied from 1.3 to 1.44 and modal values are 1.3. This is particularly notable for b928 R2 in the MBL where there were enough giant mode particles (10–40 µm) counted, showing much larger modal aspect ratios of 1.5 (median of 1.50). For flight b920 the smallest particles (0.1–0.5 µm) were more spherical (modal aspect ratios 1.0–1.2, median values 1.08 and 1.23), though b928 R2 and b934 R6 contrast with this with much greater fractions of higher aspect ratios for smaller particles (mode aspect ratios 1.2–1.4, median values of 1.27 and 1.13, a larger tail in the aspect ratio distribution). This is explained by the composition of these latter two SLRs being more strongly dominated by sea salt with a cuboid shape. Data representing the accumulation mode and full PSD strongly shadow the smallest size bin shown, since the data are dominated by smallest particles with the highest number concentrations. Our median values are slightly lower than the majority of those reported in the literature. For example, SAMUM1 values were around 1.6 for d > 0.5 µm and 1.3 for smaller particles (Kandler et al., 2009), while at Praia, Cape Verde, Kandler et al. (2011a) found values of 1.6–1.7 at d > 0.7 µm and values below 1.4 at d < 0.7 µm, and Chou et al. (2008) found median values of 1.7 during African Monsoon Multidisciplinary Analysis (AMMA). Contrastingly Rocha-Lima et al. (2018) found lower modal values of 1.3 from ground-based samples during Fennec.

In order to facilitate comparisons with model data, where typical output is in terms of dust mass loadings, Fig. 10 shows dust mass loadings from all the 31 AER-D in situ aircraft profiles, calculated from the measured size distributions and assuming a typical dust density of 2.65 g cm-3 (Tegen and Fung, 1994) and spherical particles. The intense dust profiles encountered during certain sections of flights b923/b924 are highlighted in orange, where dust was elevated further north around the Canary Islands, and in red where the dust was found at lower altitudes at around 23∘ N, since they both show notably higher dust mass loadings. Due to the presence of a strong coarse and giant mode, mass loadings are generally high and typically 300–1000 µg m-3 in the elevated SAL (Fig. 10a, black lines) when AODs were low to moderate (0.2 to 0.5) (black lines). Exceptionally high values were found during the intense dust event on 12 August 2015 when values of mass loading exceeded 1000 µg m-3 and reached a maximum value of around 4600 µg m-3. These values are of the order of a factor of 10 larger than those observed in the region previously, which measured up to a maximum size of 20 µm (Collins et al., 2000; Garrett et al., 2003).

Figure 10b shows that the mass contained within the accumulation mode (d < 2.5 µm) is around a factor of 10 lower than the total mass, indicating that sub-sampling dust properties behind size-limiting inlets significantly under-samples dust mass. Specifically, we find that on average 14 % (10th and 90th percentiles at 6 % and 28 %) of total mass is contained within the accumulation mode between altitudes of 1.5 to 4 km, values in agreement with estimates by Kok et al. (2017). Figure 10c and d further illustrate the impact of size on the mass loading by showing the fraction of mass contained at diameters greater than 5 and 20 µm (c and d respectively). These values are selected since 5 µm is the diameter at which models begin to under-represent dust mass concentration compared to observations (Kok et al., 2017), and few models represent dust particles larger than 20 µm (Huneeus et al., 2011). On average, around 60 % of the mass is found at sizes greater than 5 µm and 0 %–12 % at sizes greater than 20 µm in the SAL between altitudes of 1.5 and 4 km where most of the mass is found. Within the MBL a greater fraction of mass is found at large sizes: 70 %–80 % greater than 5 µm and 10 %–20 % greater than 20 µm. In the extreme, up to 90 % of dust mass can be found at sizes greater than 5 µm and up to 40 % at sizes greater than 20 µm.

There appears to be a trend with altitude shown in Fig. 10c: the mean mass fraction at d > 5 µm decreases steadily from 0.75 at the surface to 0.23 at 5 km altitude. A decrease is also evident in panel d with the largest fractions being found towards the bottom of the SAL (excluding the MBL). These decreases with dust mass as a function of altitude are somewhat in contrast with the homogeneous distribution of deff throughout the SAL shown in Fig. 8. This may be due to the data shown in Fig. 10 coming from profiles rather than SLRs, such that more data are available and also that although deff represents the full size distribution, as such it is relatively insensitive to smaller changes in the coarse and giant particle concentration. Either way, there is clearly evidence of coarser dust particles being more prevalent towards the bottom of the SAL (and also in the MBL), indicating deposition processes occurring.

Additionally, we have calculated dust mass path (DMP, also known as integrated column mass loading) values from the mass profiles, where DMP is the vertically integrated mass of dust per unit area (Evan et al., 2014). For AER-D low to moderate AODs, mean DMP is 1.0 g m-2 (minimum and maximum values of 0.2 and 2.4 g m-2). These values are higher than, but within, the bounds of error and variability of those derived from satellite retrievals in the same geographic region (Evan et al., 2014). However, DMPs produced by CMIP5 models are much lower (0.05 to 0.46 g m-2 with a multi-model median of 0.26 g m-2) – falling at or below the lower edge of the AER-D values, furthering the argument that models underestimate dust mass loading due to poor representation of dust coarse mode. For the intense dust event on 12 August, DMP values are extremely large, from 3.1 to 6.2 g m-2.

Figure 11 shows the size-resolved and bulk (full PSD and accumulation mode, d < 2.5 µm only) composition results for the four filter samples analysed. In general, across all the samples and all size ranges above diameters of 0.5 µm, the particles are dominated by alumino-silicates and quartz, with alumino-silicates forming over 80 % of the composition volume, with quartz typically forming around 10 % of the volume, though sometimes being up to 20 %, consistent with Price et al. (2018). Other components are generally low in volume percentage, with calcium-rich particles providing up to 15 % of volume content in some samples. Particles dominated by iron are present in low quantities (and therefore displayed separately on the right-hand axis), although their contribution is extremely important in controlling shortwave refractive indices (Sect. 3.5). Fe-rich particles are present in higher quantities in the SAL cases (0.5 %–2.5 %) compared to the MBL (0 %–0.9 %). Error bars are noticeably large for Fe-rich particles due to the relatively low counting statistics. Although there is some variation in the iron content as a function of particle size, no distinct pattern is displayed across all the samples, and the uncertainties prevent definitive conclusions from being drawn. Even when the uncertainties are reduced by looking at the full PSD vs. accumulation mode only (black circle and triangle), iron content is higher for the full PSD than the accumulation mode in the SAL for b920 R5 (1.7 % vs. 1.3 %), whereas for b934 R6 also in the SAL, iron content is higher in the accumulation mode than for the full PSD (1.3 % vs. 0.6 %). These values agree well with the range of hematite content proposed by Balkanski et al. (2007), where values span 0.9 to 2.7 %, though our values will be low-biased as we only include iron when iron was the dominant component of a particle, and it is evident that iron is present as a portion of almost all larger particles. Iron is detected across the full size range in variable amounts; thus, where absorption is increased due to large particle sizes, it will be further increased due to elevated iron content.

The size range 0.1 to 0.5 µm (shown in dark blue) displays a notably different composition, in keeping with the different size distribution displayed for the fine mode (Figs. 5 and 7) indicating a different aerosol type. Although some fraction of this fine mode is composed of dust particles (alumino-silicates and quartz), there is always a contribution from sulfate particles (10 %–40 %). Contributions from sea salt in the MBL fine mode (25 % and 55 %) are higher than in the SAL, as expected. Samples B928_R2 and B934_R6 show noticeably higher thenardite concentrations, possibly indicative of dry saline lake origins (e.g. Formenti et al., 2003) or sea salt reacted with sulfuric acid. The high contribution of sea salt in the fine mode in b928 R2 (Fig. 11b) impacts the aspect ratio distributions for this size range shown in Fig. 9b, where these small particles are much less spherical and present a rectangular shape in the SEM imagery, inferred to be a cuboid shape.

Figure 11 shows that even within the MBL, the aerosol content was dominated by dust particles in both cases examined here. Additionally, during b928 R2 (orange points in Figs. 5b, and 11b), one of the SLRs when a substantial giant mode was present, the composition data confirms that this giant mode (purple points) was composed of dust, being dominated by alumino-silicates, quartz and calcium-rich particles (most likely calcite).

Kandler et al. (2009, 2011a) show that the quartz fraction for particles increases with particle size, particularly for d > 20 µm. Our data do not show that quartz content appreciably increases with size, though this may be due to the relatively large errors on the size-resolved data, although even the bulk data for the full PSD and accumulation mode PSD with smaller error bars do not show significant differences in quartz content. It may also be due to our filter data not extending to such large size ranges (up to 200 µm) as Kandler et al. (2009).

Additionally, in contrast with Liu et al. (2018), we do not detect any black carbon on the filter samples, which they find present predominantly between sizes of 0.1 and 0.6 µm. During the analysis of 6500 particles, only one black carbon chain structure was observed. This difference is unlikely to be due to the pore size of the filter samples (0.2 and 0.4 µm) allowing small black carbon (BC) particles to pass through them, since our filters collect efficiently over a size range extending below this by diffusion and impaction (e.g. Lindsley, 2016), and the size distributions in Fig. 7 clearly show that this size range is collected. BC loadings for the ICE-D flights shown in Liu et al. (2018) are very low (0.05–1.0 µg m-3). The most likely explanation is that BC particles were not present in sufficient concentrations to be sampled by the filters.

Size-resolved and bulk complex refractive indices at 550 nm calculated from the filter samples' composition are shown in Table 7, and full spectral variability (solar and terrestrial) is shown in the Supplement, Figs. S2 and S3. Real values in Table 7 representing the FULL PSD and accumulation mode PSD are generally 1.47 to 1.49 and do not vary substantially. These values are relatively low compared to that expected for dust and may be influenced by the lower real part of illite in determining the “alumino-silicate mean” refractive index which is an average of illite and kaolinite. Other refractive indices such as those of Kandler et al. (2009, 2011a) solely apply kaolinite refractive indices and produce higher real values. The low values may also be influenced by the internal volume mixing rule applied here and are consistent with those from Formenti et al. (2014). Additionally, the percentages of kaolinite and illite by mass respectively have been shown to vary between 71.1±13.5 % and 16.4±11.8 % for dust originating from southern Mali–Algeria to 30.7±3.2 % and 54.2±6.0 % for dust from Mali–Mauritania–Western Sahara (Formenti et al., 2014). Real values are also distinctly higher for the smallest size class, though this has no impact on the bulk value due to the low volumetric contribution from these small-sized particles.

Much more variability is seen in the imaginary part, largely influenced by the iron content and its absorbing influence via hematite and goethite. For the two samples in the SAL, k is the same for both in the accumulation mode (0.0023) but increases or decreases to 0.0030 or 0.0012 when the coarser particles are included. To explain this, inspecting the size-resolved data, it can be seen that for sample b920 R5 k increases with size from 0.0020 to 0.0034 reflecting the increasing iron content with size seen in Fig. 11c. However, this is not the case for the second SAL sample (b934 R6) where iron content and k are highest (0.0023 to 0.0039) in the mid-range sizes from 0.5 to 5 µm (Fig. 11d), while lower k values (0.0003 to 0.0009) are found for d < 0.5 µm and d > 5 µm due to their lower iron content.

Imaginary parts for the MBL are much smaller than those for the SAL (0.0004–0.0005 compared to 0.0012 to 0.003), reflecting the lower iron content across all size ranges. Real parts in the MBL are not notably different to those in the SAL.

Longwave spectral refractive indices have also been calculated and are shown in Fig. S3. Here the main controlling factor is the fraction of sea salt and sulfate-rich particles, which is important for d < 0.5 µm and occasionally up to d=1 µm in the MBL, and the relative proportion of alumino-silicate and quartz is important for sizes d > 0.5 µm. The refractive indices of sea salt used here have zero absorption in the longwave spectrum, so higher salt content lowers k in the smallest size range. All samples show an increase in absorption within the atmospheric window between around 8.5 and 10 µm. The quartz absorption peak occurs at 9.4 µm while the alumino-silicate peak occurs at 9.6 µm. Thus, the relative proportions of these two minerals control the height of each of these peaks, but since no significant size-resolved change in the quartz : alumino-silicate ratio was found in Sect. 3.4, any size-resolved changes in these peaks in the longwave refractive index are negligible.

The longwave imaginary parts are substantially higher between 9 and 10 µm, by 2 to 3 times, than those of some of the literature where k is less than 1.0 (Di Biagio et al., 2017; Fouquart et al., 1987; Volz, 1973; Hess et al., 1998; Balkanski et al., 2007), although our values are similarly high to those of both Otto et al. (2009) and Formenti et al. (2014), who used similar internal mixing volume calculations. We purport this to be due to the assumption of an internally mixed dust aerosol to be inappropriate, knowing that the dust is actually externally mixed (excepting the complex nature of iron existing within mixtures), as in Formenti et al. (2014), though why this assumption produces worse results in the longwave spectrum is unclear. Other ways of calculating the refractive index based on iron being internally mixed, while quartz and alumino-silicates are externally mixed, are possible (e.g. Balkanski et al., 2007), as are other more complex ways of representing internal and inhomogeneous mixing (Lindqvist et al., 2014; Nousiainen, 2009), but they are beyond the scope of this work.

Figure 12 shows how SSA varied with height for the accumulation mode and the full size distribution. SSA was fairly constant with altitude within the SAL. Measured mean (minimum, maximum) accumulation mode SAL SSAs (black) were 0.97 (0.93, 0.98). MBL values were all greater than 0.99. Mie-calculated values of SSA for the accumulation mode (orange) agree with measurements (black) within error bounds, since here agreement in SSA is tuned by refractive index iterations. Once the full size distribution had been accounted for, calculated SSA values (green) decrease to 0.95 (0.91, 0.98) for the SAL and to 0.99 (0.97, 0.99) for the MBL. Since larger particles are more absorbing for a fixed refractive index, this decrease is expected, and the magnitude of the decrease is dependent on the amount of coarse particles and also the refractive index. In the MBL, SSA values do not decrease much since k550 is small (∼0.0001i); the addition of a coarse mode of negligible absorption makes little difference to SSA. However, in the SAL, where derived k550 varies from 0.0005 to 0.0025i, the addition of the full coarse mode causes SSA to drop noticeably.

Since it is clear from the composition results that aerosol in the SAL is dominated by dust across all sizes, applying the same k550 for the accumulation mode to the coarse particles may be appropriate. However, some variation in dust composition with size is still evident (Sect. 3.4), which is not taken into account in Fig. 12. The low k550 of 0.0001i in the MBL is representative of highly scattering sea salt or sulfates which dominate the fine mode in the MBL (Sect. 3.4), although it is clear that dust particles dominate at larger sizes in the MBL. Therefore, extending the low value of k550 to the coarse mode in the MBL is a less reasonable assumption. Therefore, it is likely that the SSA values in the MBL for the full size distribution shown in Fig. 12 are overestimates. SSA values calculated using an RI representative of dust for the coarse mode from filter samples are discussed later on.

Theoretically, the variability of the optical properties of dust in the SAL may be determined by either the dust composition, the dust size distribution or both. Other elements may also influence the optical properties, such as particle shape, roughness, hygroscopic growth and mixing, but they are not considered here as there are few observational constraints available on these properties from AER-D. In order to determine the contributions from both particle size (assessed via deff from the measured size distribution) and composition (assessed via k550 calculated from two different methods described in Sect. 2.5) to the variability of SSA, we show both deff and k550 as a function of SSA in Fig. 13, separated into the ACC PSD and FULL PSD size ranges.

It is clear that during AER-D, it was the variability of dust composition that controlled the variability of SSA rather than the variability of size distribution. In Fig. 13a for the SAL (orange) points, it can be seen that the direct observations for the accumulation mode (orange asterisks, where nephelometer and PSAP scattering and absorption are used to calculate SSA) are in agreement with Mie simulations using the ACC PSD (small orange circles). When the coarse-mode size distribution is then included (large orange circles), effective diameter increases and there is a slight decrease in SSA. Although deff does impact the magnitude of the SSA, within AER-D the shape of the PSD was relatively stable, with a relatively constant deff. This meant that the small variation in size distribution shape had minimal impact in determining the variability of SSA.

Figure 13b shows the variation in SSA as a function of composition, represented by k550. Here we show the relationship between these two variables calculated with different RI datasets: firstly for RI from the Mie scattering iterations (circles, Sect. 2.5.1) and secondly for RI calculated for filter sample composition (diamonds, Sect. 2.5.2). Optical properties are shown for the accumulation mode only (small data points), and for the FULL PSD (large data points). For dust in the SAL, Fig. 13b shows a consistent decrease in SSA with the imaginary part of the refractive index. This relationship becomes more negative when the full coarse-mode PSD (large symbols) are included because for a fixed RI, the larger particles exert more absorption. Although there is some variability between the results from different RI datasets, overall they show the same trend.

Contrastingly to Fig. 13a, Fig. 13b shows that the SSA variability was strongly influenced by the variability in composition. This is the case for both accumulation mode observations of SSA and for the full size distribution. It is not surprising that variability in k550 influences absorption and therefore SSA. However, the SSA can be influenced by several factors, including the PSD. Our aim is to investigate which factors influence the variability of the SSA. Therefore, it is notable that there is so little variation in the PSD during AER-D that the composition (or k550) is the main factor contributing to the variability of the SSA. This finding is notably the opposite of that found during Fennec, where the size distribution was the dominant controller of optical properties. Liu et al. (2018) show that hematite content is important in the ICE-D/AER-D samples as a controlling factor on optical properties. Moosmuller et al. (2012) and Caponi et al. (2017) also show dependencies of refractive index on iron content. This is consistent with our findings that the calculated refractive index from the filter samples is strongly influenced by the iron content and its absorbing properties. It appears that over the Sahara, variations in the PSD (affected by dust age) have an important impact on SSA, while over the ocean the impacts of composition (perhaps either by chemical aging or by sampling dust from different sources) become more important.

Finally, we compare optical properties calculated with composition data from the filter samples using both internal and external mixing assumptions against observations in Fig. 14 for the ACC PSD (a) and the FULL PSD (b). Observations are available only for the ACC PSD size range. Figure 14a shows that compared to the observations of SSA, the external mixing assumption provides closer agreement with the observations, confirming that the internal mixing assumption used to derive refractive indices shown in Sect. 3.5 overestimates the absorption, consistent with previous publications (Formenti et al., 2014). Figure 14b shows the same results but for the FULL PSD and also those for the iterated RI for the FULL PSD. The internal and external mixing calculations, when applied to the full size distribution, allow for the composition of larger particles being dust in the MBL, contrasting with the iterated RI method which assumes constant composition across all sizes. For the two MBL SLRs the SSAs drop from 0.99 for the ACC PSD down to 0.98 and 0.97 for external mixing when the coarse mode and its composition is accounted for – thus producing SSAs in the MBL similar to those of the overlying SAL. This is only achieved through analysis of the coarse-mode aerosol composition within the MBL. SSAs from internal mixing RIs are also 0.98 and 0.97 for the MBL SLRs, but these are much the same as those for the accumulation mode. For b934 R6 there is not much variation between the three methods, while for b 920 R5 the SSA is lower at 0.89 for internal mixing, compared to 0.96 for the other two methods. This is because b934 R6 contained more iron in the coarse mode than the other SLRs (see Fig. 11c) which had a strong impact on lowering the SSA for an internal mixing assumption. This is not reflected in the external mixing value due to the non-linearity of the scattering and absorption properties.

Campaign mean optical properties representing the full size distribution in the SAL and MBL are summarized in Table 6. In AER-D, MEC in the SAL is higher (0.27–0.35 m2 g-) compared to Fennec observations over the Sahara (0.15–0.23 m2 g-1, Ryder et al., 2013a) as a result of fewer coarse and giant particles. AER-D SAL SSA values (0.91 to 0.98, mean 0.95) are somewhat higher than those of Fennec (0.86 to 0.97) for the same RI (Ryder et al., 2013b) and closer to previously published higher values for transported dust. For example, (Haywood et al., 2003) values of 0.95–0.98 during the Saharan Dust Experiment (SHADE) and Chen et al. (2011) values of 0.97±0.02 during NASA AMMA (NAMMA). AER-D SAL asymmetry parameters are large at 0.74, unsurprising given the presence of coarse particles, contributing to forward scattering.