UMBC–LACO deployed two automated LACO aerosol sampling stations, one at each Fennec supersite. The aerosol sampling station is a system for collection of aerosol particles on filters designed and built at UMBC. This instrument has a cartridge with space for 16 filters, separated into two stages for eight fine filters and eight coarse filters. Nuclepore filters with 25 mm diameter and 5.0 and 0.4 µm pore diameters were used as coarse (first-stage) and fine (second-stage) filters, respectively, to collect the aerosol particles. Particles with aerodynamic diameters larger than 10 µm were removed by the aerodynamic impactor of the instrument inlet. This impactor has a cut efficiency of 50 % for particles with aerodynamic sizes of 10 µm in diameter and density equal 1 g cm-3 . For particles of density around 2.6 g cm-3, such as dust, this is approximately equivalent a cut size of 50 % at particle's diameter of 6.1 µm, assuming a spherical shape of the particles.

The first-stage filters adequately prevent coarse particles from passing through the pores to adhere to the second-stage filter. Thus the second-stage filter represents a fine-mode aerosol and the size distributions analyzed from the second-stage filters include only particles with diameter less than 5 µm. The coarse particles in the sample adhere to the first-stage filters, but so do many fine particles. There is overlap of size distributions of the first- and second-stage filters, causing us to identify the first-stage filter as representing a “mixed” size mode aerosol rather than a coarse mode.

Each sampling position in the cartridge is connected individually through vacuum tubes to the control system unit containing automatic valves, flow meters, pump controller, and the data acquisition system. The filters were pre-weighed and the cartridges were prepared, individually labeled, and packed at the LACO filter laboratory at UMBC to avoid in field contamination. The filter in the eighth position of each cartridge was not sampled and was used as reference blank. The airflow pumped by the sampling station through the filters was set at 4 L min-1. At the end of the campaign, cartridges containing the sampled filters were sent back to the laboratory at UMBC for detailed analysis of mass, size, and aspect ratio distribution, chemical composition, and spectral optical reflectance measurements. See Table  for deployment durations and sampling periods.

The inverse integrating nephelometer and optical reflectometer (N–OR system) were designed to make real-time measurements of the scattering and absorption coefficients of ambient aerosol particles. This instrument connects an inverse integrated nephelometer (N system) with an optical reflectometer (OR system) into a single unit that was designed, built, and tested at the LACO at UMBC.

The N system component measures the total scattering coefficient integrated over an angular range of 5–178∘. An aerodynamic impactor in the inlet of the equipment cuts off particles larger than 10 µm in diameter. The internal laser beam with wavelength of 670 nm illuminates particles entering the inlet of the instrument. A photomultiplier tube detector and a cosine diffuser are positioned perpendicular to the laser beam, aiming to maximize the scattering angle coverage of the instrument.

At the end of the N system, the OR system component measures the change of reflectance of a Nuclepore filter in real time as the particles collect on the filter and darken the surface. The OR system uses three LEDs at wavelengths 450, 530, and 640 nm to illuminate the filter consecutively and allow for derivation of absorption at these three wavelengths. This allows the simultaneous measurement of the scattering and absorption coefficients of the aerosol particles, the calculation of SSA for the 670 nm in real time (by scaling the absorption coefficient from 640 to 670 nm using our spectral measurements discussed on Sect. 4.4), and the creation of time series of these optical parameters with a temporal resolution of 45 s.

Table 1 summarizes the characteristics and sampling periods of the ground-based measurements obtained by the LACO–UMBC instruments during Fennec 2011. The sampling period corresponds to the temporal resolution of each instrument. For the LACO aerosol sampling station, the sampling period of each filter, i.e., the interval on which filters were being replaced, was on the order of hours. Therefore, the properties obtained later on from filter-based measurements are time-averaged over the sampling period. A summary of the physical quantities retrieved by each of the LACO–UMBC instruments is presented in Table .

The LACO–UMBC instruments were deployed at SS1, BBM in the heart of the central Sahara, and in a small village called Bir Moghrein approximately 290 km north of SS2, which was located in the city of Zouérat, Mauritania. The reason for the deployment at a distance from SS2 was to avoid contamination from aerosols produced from local mining operations in Zouérat. See Fig. .

SS1 is located in BBM, Algeria (21.38∘ N, 0.92∘ E; ≈ 420 m a.s.l. – above sea level). In addition to the LACO–UMBC instruments, the Fennec team deployed other instruments at this location, including an AERONET Cimel sun photometer. A detailed description of the other instruments and measurements of the Fennec campaign at SS1 is available in . Figure  shows an image of the tower where the instruments were installed. The inlets of both aerosol sampling stations and the nephelometer were positioned at a distance of 3 m from the ground. The installation of this tower close to the ONM and to the airport facilitated access to the tower for replacement of filter cartridges. Also, it allowed the operation of the aerosol sampling station in manual mode, in which the operator collected more filters during intense episodes of dust.

The location of the second automated LACO aerosol sampling station was in the remote town of Bir Moghrein (25.23∘ N, 11.62∘ W; ≈ 360 m a.s.l.). The aerosol sampling station was operational from 23 May to 26 June 2011 collecting three to four filters per day with its inlet also located 3 m above ground. Given the difficulty in accessing the Bir Moghrein site, the station was preset to automatically sample filters at the following periods of time: 07:00 to 13:00, 13:00 to 19:00 UTC, 19:00 to 21:00, and 21:00 to 07:00 UTC. No N–OR was deployed at Bir Moghrein, and therefore no high temporal resolution data were collected with the LACO–UMBC instruments at this site.

LACO aerosol sampling stations were deployed at both sites, allowing for measurements of aerosol mass concentration as function of time, with a resolution of 6 h except during intense dust episodes at SS1, when samples were collected at a higher frequency. The sampling station automatically advanced measurements from filter to filter during measurement periods defined in Sect. 3.2, accumulating aerosol mass on four different filters each day. After the end of the campaign, cartridges containing the sampled filters were sent back to UMBC. At the laboratory, each filter was post-weighed and the mass collected in each filter was obtained. The mean mass aerosol concentration for the period that each filter was sampled was obtained by dividing the sampled mass by the integrated flow of the sampling period of each filter. The temporal resolution of the mass concentration time series is nominally 6 h, based on the 6 h sampling period of each filter, and the mass concentration time series is not a real-time measurement. The sampling station filter cartridge supports two size stages for each sampling period, as defined in Sect. 2.2, and thus the mass concentration time series is available for both the aerosol loading on the coarse-pore (first-stage) and the fine-pore (second-stage) filters. The time series of integrated scattering coefficient (βsca) was measured in real time every 4 s using the N system located only at SS1.

Figure  shows the concentration (in µg m-3) from the first- (mixed) and second- (fine) stage filters for both stations in (a) Algeria and (b) Mauritania. Note that while the first-stage filter successfully prevents coarse particles (d > 5 µm) from passing through to the second stage, allowing the second-stage filter to represent a true fine-mode aerosol, both fine and coarse particles adhere to this first stage, creating a mode of mixed sized particles. This size separation of the two filters will be shown in Sect. 5. In Fig. , we see that the mass concentration at SS1 in Algeria reached levels approximately 10 times larger than in Mauritania. In Algeria, the highest peaks of mass concentration were observed on 13 and 18 June, with lesser events noted on 16, 21–22, and 29–30 June. These peaks are associated with the sudden moistening convective events described by . also associated part of the dust loading in some of these events to a breakdown of a low-level jet. On 25 June there is also a moistening event, but it does not have a corresponding peak in mass concentration as measured by the aerosol sampling station at SS1. In Mauritania, we see the distinction between the “maritime phase” with low aerosol loading that occurs from 1 to 13 June and the onset of the “heat low phase” after that period with higher aerosol loading and greater influence from the interior desert, as described by .

Figure a also shows the integrated scattering coefficient (in Mm-1) for the whole period of the campaign. In the first week, there was a long period with relatively low dust loading followed by a sequence of intense episodes of high dust concentration. On 13 June, the high concentrations of aerosol exceed the saturation limits of the nephelometer, and all the data above 3500 Mm-1 were not measured. On 22 June, the detection scale of the nephelometer was reconfigured in order to allow for higher dynamic range and prevent saturation. The period from 22 to 30 June has the optimum configuration conditions for the N–OR system. Note that the days of peak scattering coefficient at SS1 correspond to some of the same days of independently measured high mass concentration, i.e., 13, 18, 21–22, and 29–30 June. The 25 June also shows a high scattering coefficient, but that day is missing from the mass concentration time series, although 25 June, like the other observed high mass and high scattering events, follows a moistening event identified in .

It is important to note that dust events of short duration observed in the scattering coefficient are not captured by the measurements of mass concentration. This happens due to different time resolution of the instruments used for these two measurements. The mass concentration is averaged over 6 h, while the nephelometer had a time resolution of 45 s.

Simultaneous to the scattering coefficient measurements, the same dust particles that passed through the nephelometer are collected on filters and measured in real time by the reflectometer. Due to a technical problem, the reflectometer did not work properly at the beginning of the campaign. The reflectometer data presented in the next sections were obtained after 22 June, after the problem had been identified and fixed.

The mass absorption efficiency (αabs), in m2 g-1, is the parameter derived from the measured quantities of attenuated reflectance of the filter (I/I0) and aerosol mass concentration (σ), in g m-2. The relationship between the measured quantities and αabs is given by Eq. (1): σ=G2αabs-ln⁡II0b, where I is the measured reflectance and I0 is the reflectance of the clean pristine filter. The functional relationship between σ and the reflectance ratio is a power law of the logarithm, where b is an empirical power law coefficient determined to be 1.218 and G is a geometrical factor determined to be equal 1 for a large range of geometries of the setup (angle between light incidence and detection), including the one used in this work. This method was previously derived and calibrated by using Monarch 71 black carbon particles manufactured by the Cabot Corporation, it was compared and showed good agreement against other absorption techniques by and applied to volcanic ash samples . This technique is based on the assumptions that the reflectance of the filter decreases as particles are loaded on it and that the reduction of the light reflected by the filter is due to absorption only by the aerosol particles. The reflectance of the filters for the red wavelength slightly increases when the first particles are collected on their surface. We estimate that this effect is the on order of 1–2 %, and it is included in the uncertainties of the reflectance. A detailed description of this method can be found in and .

This technique is based on the assumptions that the reflectance of the filter decreases as particles are loaded on it and that the reduction of the light reflected by the filter is due to absorption only by the aerosol particles. The reflectance of the filters for the red wavelength slightly increases when the first particles are collected on their surface. We estimate that this effect is on the order of 1–2 %, and it is included in the uncertainties of the reflectance.

The absorption coefficient (βabs) in Mm-1 is αabs/ρ, where ρ is the aerosol concentration (g m-3). For real-time calculations of βabs using the reflectometer we measure the darkening of the filter as a ratio of reflectance at two points in time, t1 and t2, with I(t1) substituting for I0 of Eq. (1). The darkening of the filter is thus relative to the previous measurement and is no longer referenced to the pristine filter. Then, the power law equation given by Eq. (1) was rewritten in terms of the linear absorption coefficient βabs = αabs/ρ, where the concentration ρ = M/V (g m-3) depends on the aerosol mass M collected and the volume V of air that passed through the filter in the time interval Δt = t2 - t1. We used Δt = 7.5 min to have higher sensitivity to the change of filter reflectance. The total volume is obtained by integrating the measured flow F in time, i.e., V = F ⋅ Δt. Using the aerosol mass concentration σ as the ratio of the aerosol mass M and the sampled area of the filter Afilter, in Eq. (1), the dependence with the aerosol mass M cancels out. βabs=12-ln⁡It2It11.218⋅AfilterF⋅Δt×106

Figure  shows the results of the simultaneous measurements of scattering and absorption coefficients (βsca and βabs) at SS1 for the period of 22–30 June. Figure a shows the scattering coefficient in Mm-1 for 640 nm. This is a temporal subset of the plot in Fig. a. During this period several episodes of high concentration of dust were detected. The largest episodes of dust were seen on 24, 29, and 30 June. Scattering measurements were taken every 4 s and are shown in the plot averaged every 45 s in order to correspond to the time scale of the reflectometer. Figure b shows the reflectance of Nuclepore filters at three wavelengths normalized by the reflectance of the clean filter measured at the beginning of the sampling. Filters in the reflectometer were replaced approximately once per day. The sampling start time of each filter can be identified as the moment where the reflectance is close to 1. The slope of the curve of the reflectance is proportional to the amount of aerosol in the filter at that moment and therefore to the concentration of the particles collected on the filter at that given instant.

Figure a and c show similar trend between scattering and absorption coefficients, obtained by the nephelometer and the reflectometer, respectively. The uncertainties in the scattering measurements were estimated to be smaller than 5 % for highly scattering particles such as dust particles. The uncertainties in the absorption coefficient were estimated from the error in the reflectance, integrated flow, and size of the filter where particles were collected in the filter. These errors combined represent an uncertainty on the order of 3 % (or 2 Mm-1, whichever is higher) in the absorption coefficient.

The SSA is defined as βsca/(βsca + βabs). By obtaining simultaneous scattering and absorption coefficients at SS1, the calculation of a time series of SSA was possible at 670 nm, the wavelength measured by the nephelometer. Because βabs is obtained at 640 nm, this value of was extrapolated to 670 nm based on spectral laboratory measurements showed in Fig. a. The SSA time series are shown in Fig. d. These results show variation along this period from 0.96 to close to 1, with a mean value around 0.995, which is around 0.01 to 0.02 systematically higher than the values for Saharan dust found in the literature at this wavelength, as we discuss in Sect. . It is important to note that SSA is a size-dependent quantity, and comparisons with other measurements should take into account size differences, as discussed in Sect. .

The spectral reflectance from 350 to 2500 nm was obtained for all sampled filters relative to blank filters using a FieldSpec Pro from Analytical Spectral Device in the wavelength range of 350 to 2500 nm and a reflectance lamp from ASD Inc. The method applied in this analysis followed the same experimental procedure used in .

The reflectance of the filters collected at both stations was measured relative to a white reference. Figure  shows examples of reflectance spectra for fine and mixed modes for filters of different mass loading collected on different days in Algeria (SS1). The typical reflectance spectrum obtained for the filters in Algeria shows a sharp decrease in reflectance for wavelengths less than 650 nm. This strong spectral dependence is what causes the dust to appear brown to our eye. Some of the samples also presented a slight decrease in reflectance above 1000 nm. The ripples on the reflectance curve above 2000 nm are measurement artifacts also observed in clean filters.

The reflectance of the filters collected in Mauritania presented significant spectral variations and three groups of samples were identified based on the qualitative inspection of the shape of the reflectance curve in the visible and NIR, as shown in Fig. . Group 1 shows spectral reflectance similar to the reflectance of the samples collected in Algeria as shown in Fig. , with strongly decreased reflectance below 600 nm. Filters from the group 2 have reflectance spectra with a flat signal spanning the UV to visible wavelengths, a minimum of reflectance around 860 nm, and then a slightly increasing reflectance as a function of wavelength through the shortwave infrared. Finally, group 3 has a reflectance spectrum that seems to be a combination of groups 1 and 2.

SEM images of these three groups show that groups 2 and 3 contain extra-large particles comparatively to the sizes from group 1. While the SEM images show irregularly shaped particles, reminiscent of dust, our laboratory observations indicated that these extra-large particles have lower density compared with typical dust. Firstly, the mass of the filters from group 2 is relatively low, even though SEM images show a considerable number of particles on them. Secondly, these large particles are easily damaged by the electron beam of the microscope. Finally, the presence of the aerodynamic impactor with nominal cutoff size of 10 µm (or approximately 6.1 µm for a spherical dust particle of density 2.6 g cm-3) in the inlet should have removed most of these particles, unless they have lower inertia and lower aerodynamic sizes, i.e., lower mass density.

Thus, while the spectral reflectance, size, and density of the particles of Mauritania group 1 resemble the dust properties and the measurements from SS1, the measured particle properties of Mauritania groups 2 and 3 do not. The anomalous spectral reflectance and particle sizes of groups 2 and 3 are always linked together, meaning we do not find filters with groups 2 or 3 spectral reflectance without also finding groups 2 and 3 particle size and density properties. These particles are sparse on the surface of the filters and the complete characterization of their properties and origin would require dedicated microscopy and trajectory analysis, which are beyond the scope of this work. Therefore, for the remainder of the analysis we will focus on the properties measured and derived from Algeria and from Mauritania group 1 only.

From the spectral reflectance measurements and mass concentration applied to Eq. (), we derived the spectral mass absorption efficiency (in m2 g-1) for fine and mixed particle size distributions for SS1 and SS2 (group 1 only), as shown in Fig. . The fine and mixed size distributions correspond to the particles on the second- and first-stage filters, respectively, as discussed in Sect. 4.2. For SS1, the mass absorption efficiency of both fine and mixed modes is in good agreement up to wavelength of 600 nm. Above that, fine and mixed modes deviate from each other, with the fine mode exhibiting higher values. For SS2, the mass absorption efficiency for the fine mode is slightly higher than the mixed mode for wavelengths up to 600 nm, and both modes are compatible above that. The uncertainties of these curves are represented by the bands plotted around the central value of the mass absorption efficiency corresponding to one standard deviation.

The number, area, and volume size distributions were obtained from SEM images of the dust particles. Analysis included both the first-stage filters with pore size 5 µm and the second-stage filters with pore size 0.4 µm as shown in Sect. 3.2.

Figure a and b show the fine and mixed particle size distributions for Algeria and Mauritania (group 1), respectively, obtained by analyzing approximately 2000 particles. In this example, these distributions show considerably larger fraction of particles below 1 µm of diameter in Mauritania. Compared to Mauritania, Algeria is more centered in the Saharan desert and closer to the major sources of dust and therefore with higher fraction of fresh dust. Other studies also show a decrease in coarse-mode fraction as sampling moves towards aged dust and away from fresh dust near the major sources .

The particle size distribution obtained by SEM images is the distribution sampled on the filter and for consistency this is the size distribution used to derive the optical properties of the dust collected on the filters using post-deployment measurements in the laboratory. Note that SEM-derived size distributions are not directly compared with size distributions from optical measurements or aerodynamic sizing without adjustments that are not made in the work presented here.

The technique used for the measurements of the grain density is based on the determination of the volume of the sample using the principle of gas displacement in a device under compressions and it requires a bulk sample of at least 1–2 g of the material, as was described in . The major dust storms in Algeria caused significant uplift and deposition of dust on the surface of the instruments. Once the instruments arrived back at UMBC, dust deposited on the instrument surfaces was gently collected using a brush and sieved using a 45 µm mesh grid. The resulting bulk sample obtained had the required mass needed for the grain density measurements. The average grain density obtained for the Saharan dust from Algeria was 2.69 ± 0.12 g cm-3. Because there was not enough material from SS2 in Mauritania for a grain density analysis, we used the same grain density for the samples of the group 1 collected at the supersite in Mauritania.

The grain density for dust particles reported in the literature range from 2.1 to 2.6 g cm-3 . used 2.65 g cm-3 to parameterize dust density during Fennec's airborne measurements and these values are compatible with our values measured in the laboratory.

Now that size distribution and particle grain density have been determined, the final input needed to calculate αabs from Eq. (3) is Qabs(m,x). If we assume that the particles are spherical, we can use a Mie code to calculate Qabs(m,x). However, the aspect ratio distribution of the particles shows that they are typically non-spherical. Therefore, in addition to the Mie code, the extended-precision T-matrix code , assuming randomly oriented ellipsoidal particles, was used with a modified gamma distribution fitted to the measurements. The T-matrix code requires aspect ratio of the particles as input. For the fine mode, the value of the most probable aspect ratios used was obtained as 1.3 for both supersites from the analysis of SEM images, taken from the distribution shown in Fig. . This is for the fine mode only, as the T-matrix code does not converge for coarse particles in the wavelength range we are studying. We note that the aspect ratio of mineral dust was measured to be 1.7 during AMMA–DABEX and 1.7–1.9 from samples collected during SAMUM . In Morocco, for dust sizes with diameter > 0.5 µm the aspect ratio was 1.6 . In Cape Verde, similar aspect ratios were found .

However, in a laboratory analysis of size-separated mineral dust the aspect ratio was 1.3 for small dust particles measured during SAMUM , similar to the results of our SEM analysis of the fine-mode filters.

Finally, the imaginary part of the refractive index of the dust particles was derived using a minimization method applied for the mass absorption efficiency for each wavelength and the results are shown in Fig. . This minimization consists of finding the imaginary part of the refractive index in which the mass absorption efficiency derived from measurements of optical reflectance (Fig. ) matches the mass absorption efficiency calculated using Eq. (3). The real part of the refractive index is assumed to be a constant value of 1.56 for all wavelengths. In the calculation of the absorption efficiency Qabs(m,x), the shape of the fine particles was considered to be first spherical and then spheroidal using Mie theory and T-matrix theory, respectively. For the mixed mode, only Mie theory was used since the T-matrix algorithm did not converge for larger particles size.

Figure a shows that the imaginary part of the complex refractive index for Saharan dust from Algeria has significant spectral differences between fine and mixed mode. Both fine and mixed modes present a significant increase inversely proportional to wavelength below 600 nm. For longer wavelengths the values diverge considerably, as the imaginary part of the refractive index of the mixed fraction remains nearly constant relative to the values found for the fine fraction, which increases significantly as a function of the wavelength. Similarly, the same behavior found for the Algeria fine mode is observed for both fine and mixed mode in Mauritania. For the mixed mode in Mauritania, the mass absorption efficiency and refractive index were derived for wavelengths up to 1850 nm. Above this wavelength the minimization method did not converge within an acceptable error of 5 %. Also for Mauritania, the retrieval of the imaginary part of the refractive index using Mie theory introduced an uncertainty comparable to the uncertainties from the measurements. This is likely related to the presence of more fine particles in this supersite. A possible explanation is the fact that the absorption efficiency (Qabs) has a sharper variation for smaller size parameters, which introduce more variability in the retrieval of the mass absorption efficiency in Eq. (3).

In a similar analysis derive the spectral imaginary part of the refractive index for a variety of mineral dust samples, including samples collected during SAMUM in Morocco. We compare our retrievals with their results in Sect. 7.

Selected dust samples collected in the Sahara were submitted to EDXRF using an Epsilon 5 PanAnalytical spectrometer at the Atmospheric Physics Laboratory at University of Sao Paulo. A total of 150 samples, including first- and second-stage filters from both supersites, were randomly selected for this analysis. Figure  shows the average concentration in percentage of the total mass of the main elements measured for samples from Algeria and Mauritania.

Differences in the mean elemental composition can be seen between the supersites. Notably, SS2 in Mauritania has a higher concentration of sodium (Na) and chlorine (Cl), which suggest a “marine influence”. The Ca / Al ratio in Mauritania (0.57 and 0.64 for fine and mixed mode, respectively) is larger than in Algeria (0.25 and 0.25 for fine and mixed mode, respectively). That is in agreement with the Ca / Al ratio decreases observed in the Sahara from west to east described by . Source areas and composition has also been linked in , where they have found that (Ca + Mg) / Fe ratio is higher for sources areas coincident with SS2 in Mauritania and usually lower for sources areas coincident with SS1 in Algeria. The (Ca + Mg) / Fe ratio for Mauritania was found equal to (1.74 and 1.58 for fine and mixed mode, respectively) and (1.0 and 0.87 for fine and mixed mode, respectively) in Algeria. In addition to Na and Cl, trace elements usually related to pollutants as vanadium (V), chromium (Cr), phosphorus (P), and sulfur (S) are observed in higher concentration in the fine mode at Mauritania. The concentrations of V in Algeria and in the mixed mode in Mauritania were relatively low, on the order of 100 ppm. These levels of V are compatible with the natural abundance of this element in Earth's crust . However, the concentration in the fine mode at Mauritania was on average 3000 ppm, indicating a significant contribution of anthropogenic sources near this supersite. Differences in elemental composition are made clear by plotting the ratio of Mauritania's elemental composition relative to Algeria, as seen in Fig. a and b, and the ratio of the fine and mixed modes in these supersites (Fig. c and d).

The iron content is higher in the mixed mode, although the ratio of fine and mixed concentrations is still compatible with one from the estimated uncertainties. Despite that, if we consider only the concentration of Fe in Fig. , the mean concentration in Algeria (6–7 %) is slightly higher than in Mauritania (4.5–5.5 %).

Collocated measurements performed by an AERONET Cimel sun photometer at SS1 in Algeria allowed us to compare our local ground-based measurements and derivations with those obtained from total column measurements. For example, the time series of the scattering coefficient presented in Sect. 3.2 was compared with AERONET (level 2.0) AOT for the same period (Fig. ). In Fig. a, based on , , and , we assume a 5 km deep planetary boundary layer (PBL) with a constant vertical profile of dust in order to match the units with the scattering coefficient measured by the nephelometer. We note that the clear conditions observed during the first days of the experiment are also apparent in the AERONET data. The AOT measurements are higher after 13 June, but AERONET total column measurements do not necessarily follow the fine details of the ground level observations, nor do the AOT measurements follow the full magnitude of large events. In some cases, this could be because cold pools often arrive at night , when the sun photometer is not measuring. In addition, AERONET does not report level 2.0 data during some of the major dust storm events, likely due to its cloud screening process. Figure b shows that the correlation between ground measurements and the total column is better when the scattering coefficient is below ≈ 400 Mm-1. During intense events of dust storms when the scattering coefficients reach higher values, ground-based and total atmospheric column measurements do not maintain the same correlation, as the heavy dust loads occur during haboobs or low-level-jet breakdown and are not expected to occupy the full 5 km deep layer of the well-mixed late afternoon PBL . For example, it can be seen in that during fresh dust events, as are likely to be dominant during the high scattering periods shown here, the vertical profile of dust is strongly dominated by loadings in the bottom 1–2 km of the atmosphere.

The effective imaginary part of the refractive index derived from the mass absorption efficiency measurements (Fig. a) were compared with AERONET retrievals for SS1 in Algeria retrieved during the Fennec campaign. Figure a shows our imaginary refractive index for fine and mixed mode where the dashed lines indicate the AERONET wavelengths and Fig. b shows the AERONET daily average of the imaginary refractive index for all days during the campaign, when available.

Notably, AERONET retrievals in Fig. b show a higher imaginary part of refractive index on 16 and 26 June, with a slightly increasing trend above 650 nm. AERONET retrievals of size distributions give higher concentrations of fine particles for these days, which is also observed in our in situ data in the time series of the fine-mode mass fraction obtained by the ratio of the mass collected on the fine and the total (fine + mixed) modes, as shown in Fig. .

The very high concentrations of fine-mode particles on these days indicate that fine particles dominated the AERONET retrieval of refractive index of the total column of aerosols. This agrees with the different spectral signatures we observe in our derivations of the refractive index in the fine and mixed modes at SS1 in Algeria. The comparison of our refractive index with AERONET retrievals at SS2 in Mauritania was not possible, because AERONET does not have measurements nearby. The closest AERONET to our SS2 station was at Zouérat, 290 km away. It is interesting to note that our retrievals of refractive index for the fine and the mixed modes from SS2 in Mauritania follow the same spectral dependence as the fine mode in Algeria, as seen in Fig. b. In addition to that, it is important to note that the fraction of fine particles in the mixed mode in SS2 is much larger than in Algeria, as seen in the particle number distributions in the top panels of Fig. a and b and also in the mean mass concentration shown in Fig. a and b. This dominance of fine particles in the mixed mode may explain why we found the same spectral dependence of the refractive index in both fine and mixed modes in Mauritania.