In Fig. 1 the location of the study sites and North African dust sources are displayed. The dust-collecting buoy Carmen (∼ 21∘15′ N, ∼ 20∘56′ W) and the sediment trap mooring site CB (∼ 21∘16′ N, ∼ 20∘48′ W) are virtually at the same position ∼ 200 nautical miles off Cap Blanc. The sediment trap station CBi (∼ 20∘45′ N, ∼ 18∘42′ W) is located ∼ 80 nautical miles off Cap Blanc. The continental dust collector Iwik (∼ 19∘53′ N, ∼ 16∘18′ W) and the meteorological station Arkeiss (∼ 20∘7′ N, ∼ 16∘15′ W) are located in a major potential dust source area (PSA 2) in the Parc National de Banc d'Arguin (PNBA) near Iwik and near Arkeiss in Mauritania. A further meteorological station is positioned in the PSA 2 in Nouâdhibou (∼ 20∘55′ N, ∼ 17∘1′ W) in Western Sahara.

The major PSA of northern African dust are summarized in a review by Scheuvens et al. (2013). Predominant dust transport towards western Africa and off the Atlantic Ocean occurs from the foothills of the Atlas mountains, Western Sahara and western Mauritania (PSA 2), southern Algeria and northern Mali (PSA 3), and western Chad, including the Bodélé depression (PSA 5) (Scheuvens et al., 2013). In contrast, dust sourced from Tunisia and northern Algeria (PSA 1) is transported predominantly to the western Mediterranean and western Europe (Stuut et al., 2009). Central Libya (PSA 4) is the most important region for dust transport to the eastern Mediterranean (Scheuvens et al., 2013).

In the following, the lithology of the geological provinces that underlie the major PSAs is outlined (Fig. 1).

The PSA 1 is underlain by the eastern Atlas chain and the northern Grand Erg/Ahnet and Ghadames basins. The outcrops in the Atlas uplift are composed of, e.g., limestones, sandstones and evaporites (Piqué, 2001). The thick strata overlying the northern Ahnet and Ghadames basins consist of, e.g., sandstones and mudstones (Selley, 1997b).

The PSA 2 is underlain by the Reguibat Shield, the Mauritanides and the Senegal–Mauritania, Aaiún-Tarfaya, Tindouf and Taoudeni basins. The western part of the Reguibat Shield is dominated by granitic rocks, while the eastern part is dominated by metamorphic and granitic rocks (Schofield et al., 2006, and references therein). West of the Reguibat Shield, the Mauritanides consist of a metamorphic belt and ophiolite (Villeneuve, 2005). West of the Taoudeni Basin, the Mauritanides are characterized by granites, quartzites and strongly metamorphosed rocks (Villeneuve, 2005). While the Aaiún–Tarfaya Basin features outcrops with dolomites and limestones, the Senegal–Mauritania Basin is characterized by very few carbonate deposits (Bosse and Gwosdz, 1996). The Tindouf Basin is characterized by mainly sandy deposits (Selley, 1997c, b). The local soils surrounding the dust collector site Iwik are composed of sandy deposits often rich in fossil shells and partly cemented by lime (Einsele et al., 1974).

The PSA 3 is underlain by the western Hoggar and parts of the Ahnet, Taoudeni and Iullemmeden basins. The Pharusian belt located in the western Hoggar is characterized by Eburnean granulites, gneiss, greywackes and magmatic rocks (Boullier et al., 1991). In the southern Ahnet Basin sandstone strata crop out. On the eastern edge of the Taoudeni Basin outcropping sediments are characterized by conglomerates, sandstones and limestones (Bertrand-Sarfati et al., 1991). The outcrops of the Iullemmeden Basin are composed of, e.g., sandstones, carbonaceous shale, laterites and massive clays (Kogbe, 1973).

The PSA 4 is underlain by parts of the Fezzan and Nubian uplifts and the Sirte and Murzuq basins. The eastern Fezzan uplift consists of ocean island basalts (Cvetković et al., 2010; Abdel-Karim et al., 2013), while sediments outcropping in the northern Nubian uplift is composed of, e.g., sandstones, limestones and gypsiferous horizons (El Makkrouf, 1988). The southern Sirte Basin is covered by sands, gravel and sand seas (Selley, 1997a). Outcrops of the eastern Murzuq Basin are composed of marine limestones and alluvial sandstones (Selley, 1997b, c).

The PSA 5 is underlain by the Chad Basin. During the Holocene, the Chad Basin was filled with fine-grained particles from the drainage of the Tibesti mountains to the north (Prospero et al., 2002). Hence, the sediments that outcrop in the central Chad Basin are characterized by fluvial and alluvial sediments such as laminated diatomites, pelites and coastal sand ridges (Schuster et al., 2009).

Saharan dust emission, transport and deposition are related to seasonal variations in atmospheric circulation (Knippertz and Todd, 2012). The intertropical convergence zone (ITCZ) shifts meridionally from ∼ 12∘ N during boreal winter to ∼ 21∘ N during boreal summer, resulting in a seasonal change in rainfall and winds over the African continent (Nicholson, 2009).

During summer, continental rainfall is most intense and the rain belt is positioned near ∼ 10∘ N with smaller amounts of rainfall near ∼ 21∘ N. Dust emission is driven by low level jets, “haboobs”, African easterly waves (AEWs) and high surface winds associated with the Saharan heat low (Knippertz and Todd, 2012). Low-level N trade winds blow and transport dust in coastal Mauritania year-round (National Geospatial-Intelligence Agency, 2006). Saharan dust is transported on- and offshore within the Saharan air layer (SAL) at an altitude of about 3 km (Diaz et al., 1976; Carlson and Prospero, 1972; Prospero and Carlson, 1972, 1970).

During winter, dust emission is driven by the breakdown of nocturnal low-level jets after sunrise, increased surges in Harmattan winds and microscale dust devils and dust plumes (Knippertz and Todd, 2012; Koch and Renno, 2005). Dust is transported within the low-level NE and E trade winds to coastal Mauritania (Dobson, 1781) and also offshore to the sediment trap mooring sites (Stuut et al., 2005).

The surface water circulation off Cap Blanc is influenced by the southward-flowing Canary Current (CC) and the poleward-flowing coastal countercurrent or Mauritania Current (Fig. 1). Underneath, the undercurrent is flowing poleward in water depths down to 1000 m (Fig. 1). The undercurrent flows along the continental slope and transports water masses originating from ∼ 5 to 10∘ N to latitudes up to 26∘ N. The poleward-flowing South Atlantic Central Water (SACW) and the southward-flowing North Atlantic Central Water (NACW) are situated below the countercurrent and meet off Cap Blanc (Mittelstaedt, 1991). The study area is positioned in a zone of permanent annual upwelling of subsurface water masses (Cropper et al., 2014). The NACW and SACW may be upwelled and mixed laterally off Cap Blanc (Meunier et al., 2012). The permanent annual upwelling of nutrient-rich subsurface waters results in high phytoplankton concentrations off Cap Blanc (Van Camp et al., 1991). As a result, the surface waters are rich in organic detritus, usually referred to as “marine snow”, and faecal pellets which are produced by marine zooplankton (Iversen et al., 2010).

Individual Saharan dust particles which settle at the ocean surface hardly settle in the deep sea. Instead, fine dust particles can be transferred from the ocean surface to the deep sea by being incorporated into marine snow aggregates and faecal pellets (Ternon et al., 2010). The aggregate formation and ballasting of marine snow aggregates and faecal pellets with marine carbonate and opal as well as with Saharan dust particles results in anomalously high sinking velocities (Iversen and Robert, 2015; Fischer and Karakas, 2009; Iversen and Ploug, 2010; Iversen et al., 2010; Ploug et al., 2008b). Dust-loaded particles that sink into the deeper water column are assumed to have a mean settling speed of ∼ 240 m d-1 at site CB (Fischer and Karakas, 2009).

Saharan dust was collected in the ocean using marine sediment traps of the type Kiel (model SMT-234/243), which are conical with an opening of 0.5 m2 (Fig. 2). The principle of particle collection is much the same as described by Van der Does et al. (2016) and Korte et al. (2017). At the top of the opening, a honeycomb grid is installed to prevent large swimmers (> 1 cm) from entering the trap. The sediment traps were equipped with 20 sample cups which rotated according to a preprogrammed sampling interval (Fischer and Wefer, 1991). The sampling interval was chosen depending on the timing of the ship expeditions.

The sampling intervals were synchronized between the two sites. The intervals ranged from 9.5 to 26 days (Table 2). Deployment and recovery of the sediment trap samples was performed during the RV Poseidon expeditions POS445 (Fischer et al., 2013), POS464 (Fischer et al., 2014) and POS481 (Fischer et al., 2015) (Table 2). The working steps related to the trap deployment and treatment are described in Fischer and Wefer (1991). In order to prevent the outflow of water from the cups during sampling, each sampling cup was filled with 20 mL of filtered (< 0.2 µm) seawater with a salinity of 40 ‰. To produce seawater with a salinity of 40 ‰, 100 g NaCl Suprapur was added to 1 L of filtered seawater. Microbial and zooplankton activity was inhibited inside the trap samples by adding 1 mL of a saturated solution of the biocide HgCl2 per 100 mL of seawater. After recovery, swimmers < 1 cm were removed from the samples by sieving each sample through a 1 mm mesh. A McLane rotary liquid splitter was used to split the < 1 mm fraction of each sample into five equal aliquots.

The samples of two sediment trap deployments during 2013–2015 of the sediment trap mooring stations CB and CBi were chosen for grain-size analyses (Table 2). The upper traps sampled at an average water depth of ∼ 1300 m and the lower trap sampled at a water depth of ∼ 3600 m (Table 2). Dust which settles at the ocean surface is advected by ocean currents while settling in the water column. As a result, particles that settle in an area of ∼ 40×40 km2 in the ocean surface above the traps may be collected at a water depth of ∼ 1300 m (Friese et al., 2016). Two winter and two summer samples were chosen for X-ray diffraction (XRD) measurements (Table 3).

Saharan dust was collected on land near Iwik, Mauritania, with a passive dust sampler consisting of two masts (1 and 2) with two sets of five air sampling bottles each (A and B, Fig. 2). The dust sampling bottles are referred to as modified Wilson and Cooke (MWAC) samplers (Mendez et al., 2011; Wilson and Cooke, 1980) and consist of a closed Polyethylene bottle through which the wind can pass via two glass tubes with 8 mm openings. Thus, a big difference between the traps and the MWAC collectors is the much smaller collection area (44 mm2) of the MWAC collectors. The MWAC dust sampler was chosen because it is one of the most common (Zobeck et al., 2003) and most efficient dust samplers (Goossens and Offer, 2000). The sampling bottles were mounted horizontally at five different heights.

The samples collected in 2013–2015 were chosen for subsequent flux and grain-size analyses (Table 4). Saltating dust particles may be collected in the lower sampling bottles at 90 cm. However, the aim was to analyse dust transported in suspension to enable a better comparison between the continental and marine sites. Therefore, the highest sampling bottles attached to the mast at 2.90 m height were used for microscope, flux and grain-size analysis (Table 4). One series of bottles (series B2) on mast 2 were analysed with the microscope. The other three replicate samples (bottles A1 and B1 of mast 1, bottles A2 of mast 2) were analysed for flux and grain-size analysis. Of the three replicate samples, the sample with the highest mass was chosen for the interpretation of the flux and grain-size data because this bottle was assumed to have sampled most efficiently. Three samples mounted at a height of 2.40 m of mast 2 were chosen to test the effect of the chemical pretreatments that we do to isolate the terrigenous fraction from marine sediments on the resulting grain-size distributions (Fig. 2). Two winter and two summer samples that contained enough material were chosen for XRD measurements (Table 3).

Furthermore, dust was sampled with a MWAC dust sampler mounted on the mast of buoy Carmen, at about 2 m above the sea surface (Stuut et al., 2015). The masts of the buoy Carmen and of the Iwik dust sampler were aligned with the ambient wind direction via a wind vane (Fig. 2). This MWAC dust sample was also analysed for grain-size distribution.

The MWAC samples chosen for microscopic investigation were analysed with a Leica M165 C microscope. Microscope pictures were taken using a Leica DFC420 camera attached to the microscope. The software Leica application suite 3.8 was used for taking the pictures.

One-fifth splits of the sediment trap samples were analysed for dust fluxes and the bulk components following the method presented in Fischer and Wefer (1991). The lithogenic flux (mg m-2 d-1) was estimated according to Eq. (1): lithogenic material=dust=total mass-carbonate-opal-2×Corg. Organic carbon was measured after the removal of carbonate with 2N HCl using a CHN analyser (HERAEUS). Total carbon was estimated by combustion without pretreatment. Carbonate was determined according to Eq. (2): carbonate=total carbon-organic carbon. Biogenic opal was determined with a sequential leaching technique (Müller and Schneider, 1993).

The MWAC samples chosen for dust flux analyses were weighed on a Mettler-Toledo AT261 Delta Range balance with a precision of 0.0001 g. Mean atmospheric dust concentrations were estimated as Eq. (3): DL=MAR(v⋅A)⋅1η, where DL is the mean dust concentration (µg m-3), MAR is the mass accumulation rate (µg s-1), v is the mean wind speed per sampling month (m s-1), A is the cross-sectional area of the inlet tube of the MWAC sampler (m2) and η is the estimated sampling efficiency of MWAC bottles. A sampling efficiency of 90 % was assumed based on an efficiency study of Goossens and Offer (2000). Mean horizontal dust fluxes were calculated according to Eq. (4): Fh=MARA⋅1η, where Fh is the horizontal dust flux (mg m-2 d-1), MAR is the mass accumulation rate (mg d-1), A is the cross-sectional area of the inlet tube of the MWAC sampler (m2) and η is the estimated sampling efficiency of MWAC bottles.

A 1/25 split of the marine sediment trap samples was analysed for particle size of the terrigenous fraction. The samples were pretreated before measurement in order to isolate this fraction (see also Filipsson et al., 2011; Friese et al., 2016; Meyer et al., 2013; Stuut, 2001, for methodology) with the following steps: (1) removal of organic matter – addition of 10 mL of H2O2 (35 %) to the sediment sample and subsequent boiling until the reaction stops; (2) removal of calcium carbonate – addition of 10 mL HCl (10 %) to the sediment sample and subsequent boiling for exactly 1 min; and (3) removal of biogenic silica – adding 6 g of NaOH pellets to the sediment sample and subsequent boiling for 10 min. Before particle-size analysis, 10 drops of Na4P2O7 ⚫ 10H2O were added to each sample to assure the full disaggregation of the particles. The pretreatment of the MWAC samples differed from the pretreatment of the sediment trap samples as, obviously, these samples did not contain any biogenic material originating from marine plankton. Further, the disaggregation of particles needed to be kept to a minimum to allow for the study of dust transport processes, the so-called “minimally dispersed” aeolian fraction (McTainsh et al., 1997). Therefore, the MWAC samples were solely pretreated with three drops of Na4P2O7 ⚫ 10H2O before analysis. The marine sediment trap samples as well as the MWAC samples were analysed with the laser particle sizer Beckmann Coulter LS13320 at NIOZ using a micro liquid module (MLM). This instrument allows quick, accurate and precise data acquisition of large size intervals (Bloemsma et al., 2012). An analytical error of ±1.26 µm (±4.00 %) was considered for the measurements (Friese et al., 2016).

To investigate the comparability of the MWAC samples with the oceanic sediment trap samples, the particle-size distribution of the MWAC sample attached to buoy Carmen was compared to the averaged particle-size distributions of the upper and lower-trap series at site CB (Fig. 3a). The grain-size distribution of the MWAC sample was comparable to both sediment trap time series even though the sampling time period was different. To ensure that the pretreatment steps of the traps did not influence the terrigenous fraction itself, tests were done in which the on-land MWAC samples were exposed to the same pretreatment steps as the marine samples (Fig. 3b). One spring sample has been measured with and without a chemical pretreatment. Two autumn dust samples were obtained from the same height and mast and sampling interval, but from different bottles (A and B), and were measured with and without pretreatment. Figure 3b indicates that a pretreatment of the Iwik dust samples did not alter the particle distributions of the samples significantly. Further, the particle-size distribution of dust sampled with different bottles is comparable.

Two winter and two summer samples of the MWAC dust collector and the sediment trap series CBi were chosen for XRD analysis (Table 3). X-ray diffraction pattern analyses were carried out in the laboratory of the research group Crystallography (University of Bremen, Central Laboratory for Crystallography and Applied Material Sciences, ZEKAM, Dept. of Geosciences).

Due to the small amount of material in the available dust samples (< 100 mg), the preparation for the measurement was done by pipetting a demi-water-sample mixture onto glass slides. Thorough preparation commonly increases reproducibility of the results; however, the standard deviation given by Moore and Reynolds (1989) of ±5 % can be considered as a general guideline for mineral groups with > 20 % clay fraction. In addition, the determination of well-crystallized minerals like quartz, calcite or aragonite can be done with better standard deviations (Tucker, 1988; Vogt et al., 2002). The X-ray diffraction was measured on a Philips X'Pert Pro multipurpose diffractometer equipped with a Cu tube (kα 1.541, 45 kV, 40 mA), a fixed divergence slit of 1/4∘, a secondary Ni filter and the X'Celerator detector system. The measurements were carried out as a continuous scan from 3 to 85∘2θ, with a calculated step size of 0.016∘2θ (calculated time per step was 100 s). Mineral identification was accomplished using the Philips software X'Pert HighScore^™, which, besides the mineral identification, can give a semi-quantitative value for each identified mineral on the basis of relative intensity ratio (RIR) values. The RIR values are calculated as the ratio of the intensity of the most intense reflex of a specific mineral phase to the intensity of the most intense reflex of pure corundum (I / Ic) referring to the “matrix-flushing method” after Chung (1974). Unfortunately RIR values are sparse for clay minerals and long-chain organic materials hampered the quantification of our samples.

The obtained flux and size data were compared to nearby meteorological data (wind speed, wind direction and precipitation).

Wind direction, wind speed and precipitation data with a 20 min resolution were gathered for the sampling site CB (21∘17′–21∘12′ N, 20∘56′–20∘54′ W) during the buoy Carmen's deployments from November 2013 to September 2015 with a Vaisala WXT520 meteorology sensor. The size of the dataset was reduced by calculating 4 h averages. Moreover, wind direction and wind speed data with a resolution of 5 min to 1 h were gathered during sampling at site Iwik (19∘53.1′ N, 16∘17.6′ W) from January 2013 to January 2015 with a Davis 6250 Vantage Vue meteorology sensor. The size of the dataset was reduced by calculating 1 h averages. Further hourly precipitation data were gathered from the station Arkeiss (20∘7′ N, -16∘15′ W) from December 2013 to March 2015 with another Davis 6250 Vantage Vue meteorology sensor. Continental hourly wind direction and wind speed data were acquired online for the Nouâdhibou meteorological station (20∘55′ N, 17∘1′ W) from the Cedar Lake Ventures website (https://weatherspark.com).

Local daily precipitation data (TRMM 3B42 dataset, 0.25∘ spatial resolution) were derived from the Giovanni online data system, developed and maintained by the NASA GES DISC (https://giovanni.gsfc.nasa.gov). Daily precipitation data were downloaded as area-averages around CBi (20∘58′–20∘34′ N, 18∘56–18∘32′ W), Iwik (19∘41′–20∘5′ N, 16∘29′–16∘5′ W), CB/Carmen (21∘5′–21∘29′ N, 21∘2′–20∘38′ W) and Arkeiss (20∘19′–19∘55′ N, 16∘28′–16∘4′ W) according to the assumed catchment area of the upper trap (∼ 40×40 km2).

Maps of six-hourly mean surface wind vectors and speed (20th century reanalysis V2c dataset) were provided by the NOAA/OAR/ESRL PSD (Boulder, Colorado, USA) and downloaded from their website (http://www.esrl.noaa.gov/psd/).

The mapping software ArcMap version 10.3.1 was used to analyse the source regions of the dust samples investigated for mineralogical composition. A map was created with 4-day back trajectories for days with a dust storm event as depicted on satellite images. In addition, the African surface lithology was included in the map and soils rich in the minerals calcite, kaolinite and chlorite were marked.

Satellite quasi-true colour RGB images (MODIS dataset) were retrieved from the NASA Worldview website (https://worldview.earthdata.nasa.gov).

Four-day back trajectories at altitudes of 10 (following Stuut et al., 2005), 100, 3000, 4500 (following Skonieczny et al., 2013) and 5500 m were calculated ending at the dust collector site Iwik (19∘52′ N, 16∘17′ W) and at the proximal marine trap site CBi (20∘46′, 18∘44′ W) using the Hybrid Single Particle Langrangian Integrated Trajectory (HYSPLIT) model (Stein et al., 2015) and the reanalysis dataset (2.5∘ spatial resolution) on the NOAA website (http://ready.arl.noaa.gov).

An ArcGIS layer file of the African surface lithology (new_af_lithology_w_glbcvr_waterbdy_90m_dd84_final.lyr) was downloaded from the US Geological survey (USGS) website (http://rmgsc.cr.usgs.gov).

An ArcGIS shape file of the African soils (DSMW.shp) was downloaded from the website of the food and agriculture organization of the United Nations (FAO) (http://www.fao.org). The mean percentages of calcite (8.9 %), chlorite (4.1 %) and kaolinite (29 %) in the clay fraction of Saharan soils in general and for each soil type is given by Journet et al. (2014). Soils with larger percentages of calcite, chlorite or kaolinite in the clay fraction than the average percentages were marked in the ArcGIS map.

In Fig. 4 the meteorological data of sites Carmen (CB), CBi, Iwik, Arkeiss and Nouâdhibou during 2013 to 2014 are presented (see Fig. 4a for location of the sites). The rainfall frequency is given in Fig. 4b for each site. The number of rainfall events were calculated regarding the TRMM stations for precipitation rates > 1 mm d-1 because smaller precipitation amounts which were detected by the satellite may not actually reach the ground. Regarding the surface stations Carmen and Arkeiss, a threshold of > 0.2 mm d-1 was used in order to exclude events which may be related to anomalously high moisture instead of rainfall.

According to the TRMM satellite product the annual precipitation frequency was larger on the shoreline (station Arkeiss and Iwik) than offshore (station CBi and Carmen) (Fig. 4b). This may be explained by a decrease in atmospheric water vapour content due to precipitation when the winds move westward. Moreover, the TRMM satellite product indicated larger rainfall frequencies during the summer season compared to the winter season regarding the stations Carmen, CBi, Iwik and Arkeiss. Larger summer rainfall frequencies can be explained by the summer northward shift of the ITCZ to ∼ 21∘ N resulting in more frequent moist convection and rainfall in the study area.

The annual rainfall frequency at site Arkeiss and the summer rainfall frequencies at sites Arkeiss and Carmen compare quite well between the sensors and the TRMM observations. However, the spatial and seasonal trends observed by the TRMM data were not supported by the sensor on buoy Carmen and by the surface station in Arkeiss. The larger annual and winter rainfall frequency recorded with the sensor on buoy Carmen may be related to water emission from the ocean surface during time periods with strong surface winds. Further, disagreements between the surface stations and the TRMM stations may be caused by the local signal recorded by the respective rain sensor. A larger number of rain sensors would most likely improve the comparability to the TRMM data.

The wind direction and speed for the surface stations Carmen, Nouâdhibou and Iwik are displayed in Fig. 4c. The annual average surface wind velocity was at a maximum offshore at buoy site Carmen (CB) with ∼ 8 m s-1. The buoy recorded a larger average wind velocity during winter than during summer, which is consistent with this season being dominated by the trades. On the shoreline, the average wind velocity was slightly larger during summer than during winter. The predominant annual wind direction was NE at sites Carmen and Iwik, while predominant NW winds were recorded for site Nouâdhibou. The wind direction changed from predominantly NE during winter to a predominantly NNE direction during summer at site Carmen. A similar but less pronounced seasonal trend can be observed for the continental site Iwik. In Nouâdhibou, the predominant winter wind direction is NNW, switching to a predominantly NW wind direction during summer. Obviously, with winds originating from the open ocean, not a lot of dust is anticipated. Therefore, we interpret these wind directions as being very local and caused by the shape of the peninsula of Cap Blanc.

In Fig. 5 the results of the microscopy investigation of the Iwik 2013 time series are presented. In general, the majority of the particles consisted of angular and moderately spherical quartz grains with a diameter of ∼ 50 µm (Fig. 5a, b). A small percentage of large platy minerals with a diameter of ∼ 200 µm was found in all samples (Fig. 5b). Large quartz grains with a diameter of ∼ 150 to 200 µm were detected in 45 % of the samples. An anomalously high percentage of subangular and moderately spherical quartz grains with an average diameter of ∼ 200 µm was observed in one summer sample (Fig. 5c). Aggregated grains occurred in all samples. However, the percentage and size of the aggregates as well as the size of the aggregated grains differed from sample to sample. Usually, the size of the aggregated grains was ∼ 50 µm (Fig. 5a). Two samples were characterized by aggregates composed of particles with a smaller size of ∼ 20 µm (Fig. 5d).

In Table 5 the average dust fluxes are given for the sampling sites Iwik, CBi and CB. The dust concentrations at site Iwik were determined based on the measured wind speed of the meteorological sensor attached to the sampling mast. For four samples no wind data were available due to a failure of the instrument. For these samples a wind velocity was assumed based on the seasonal averages calculated from the available wind data of the meteorology sensor in Iwik (Fig. 4c). The annual average horizontal dust fluxes at site Iwik were of the same order of magnitude during 2013 and 2014. The annual average dust fluxes decreased from the on-land site Iwik towards the proximal site CBi and the distal site CB. At site Iwik the average dust concentration was at a maximum during spring and winter 2013 and 2014, with 393 and 341 µg m-3, respectively, and at a minimum in autumn 2013 and 2014, with 48 and 68 µg m-3, respectively. The dust fluxes generally decreased with collection height in the mast between 90 and 290 cm (not shown).

The statistical values of the measured grain-size distributions for the stations CB, CBi and Iwik are given in Table 5. In addition, the measured grain-size distributions for the time series of the stations CB, CBi and Iwik are displayed in Fig. 6. In Fig. 6a the average grain-size distribution for the samples of each of the three stations for the year 2013 are given. The maximum measured particle size decreased from ∼ 223 µm on land at site Iwik to ∼ 169 µm at the proximal site CBi and ∼ 140 µm at the distal site CB (Fig. 6a). In addition, the average modal grain size decreased from ∼ 48 µm at site Iwik to 16 µm at site CB (Table 5). Bimodal grain-size distributions were encountered for 23 % of the CBi 11–12 samples, 13 % of the Iwik 13–14 samples and none of the CB 24 samples. The three bimodal distributions of the Iwik 13–14 time series were characterized by an additional smaller coarse-mode population peaking at ∼ 16 µm besides the more pronounced and variable larger coarse-mode population peaking at ∼ 42 to 55 µm. The three Iwik dust samples characterized by a smaller coarse-mode population were collected during spring, summer (Fig. 6c) and autumn. The eight bimodal grain-size distributions of the CBi 11–12 time series were characterized by a variable, larger coarse-mode population at ∼ 25 to 35 µm and a variable, smaller coarse-mode population at ∼ 6 to 16 µm. The bimodal distributions were recorded for three winter, three summer (Fig. 6b, c), one spring and one autumn sample. The annual average PM10 percentage was about 2 times larger at the offshore sampling sites CB and CBi compared to the onshore sampling site Iwik. On average, about one-third of the dust mass sampled at sites CB and CBi was composed of particles smaller than 10 µm. The average geometric standard deviation of the CB samples was smaller, with 2.6 µm compared to the Iwik and CBi time series with 3.1 µm (Table 5). The lowest average mean / mode ratio was recorded for the CBi time series with ∼ 0.5 (Table 5).

In Fig. 6b–c the measured grain-size distributions for winter and summer samples are displayed. The averaged modal grain size for the summer samples was coarser grained compared to the winter samples of the respective grain-size time series (Table 5). The seasonality in modal grain size was largest for the CBi 11 upper-trap series of the year 2013 with a difference of ∼ 12 µm (Table 5). On average, the PM10 percentage was larger for dust sampled during summer compared to winter at sites Iwik and CBi. An opposite, however less pronounced, seasonal trend could be observed for site CB with generally larger PM10 percentages during winter compared to summer. The average standard deviation was larger and the average mean / mode ratio was smaller in the summer samples compared to the winter samples regarding sites Iwik and CBi (Table 5). This seasonal trend was not observed in the CB 24 upper samples (Table 5).

In Fig. 7a–c the results of the correlation between the characteristics of the dust sampled on land and the local meteorological data are presented. In Fig. 7a the particle sizes were correlated to the surface wind speed data (N=13 samples). A correlation above a coefficient of determination (R2) of 0.3 was considered significant at the 95 % confidence level for two-tailed probabilities. The modal particle size of the Iwik samples showed a positive linear correlation with the daily wind speed events with R2=0.5, which is significant at the 99.31 % confidence level. The correlation was only evident when using a threshold for wind events of 3.5 to 5.5 m s-1 and was best for a threshold of 5 m s-1. A better positive linear correlation was obtained when excluding the spring sample resulting in R2=0.7, which is significant at the 99.96 % confidence level.

In Fig. 7b the dust fluxes were correlated to the surface wind speed data (N=10 samples). A correlation above R2=0.4 was considered significant at the 95 % confidence level for two-tailed probabilities. The horizontal dust flux of the Iwik samples correlated positively to the daily wind speed events during the sampling interval with R2=0.7, which is significant at the 99.75 % confidence level. The correlation was only evident when using a threshold for wind events of 6.5 to 7 m s-1 and was best for a threshold of 6.5 m s-1. Moreover, a significant linear correlation with R2=0.6 was observed at the 99.15 % confidence level between the dust fluxes and the mean wind strengths during the sampling intervals (not shown).

In Fig. 7c the particle size of the Iwik summer samples was correlated to the local TRMM precipitation data (N=6 samples). In this case a correlation above R2=0.7 was considered significant at the 95 % confidence level for two-tailed probabilities. A good linear negative correlation with R2=0.9 was observed, which is significant at the 99.78 % confidence level.

In Table 6 the mineralogical composition averaged over all eight samples, over the four Iwik samples and over the four CBi samples is given. All dust samples contained the minerals quartz and mica. Further minerals that occurred with significant quantities but which were not present in all dust samples were feldspar, amphibole, zeolite, chlorite and palygorskite. Calcite, dolomite, gibbsite, kaolinite, smectite, sepiolite, fluellite, anhydrite, rutile and serpentine occurred only in some samples, resulting in a low average abundance ≤ 1 %. However, we argue that these minerals can be used as dust source indicators because of (1) the characteristic distribution of gibbsite, kaolinite, smectite and sepiolite in North Africa according to different weathering regimes (Biscaye, 1964) and (2) the characteristic occurrence of fluellite, anhydrite, rutile and serpentine according to outcropping rock type (Deer et al., 1992). Further minerals that occur in low abundances (≤ 3 %) were summarized as “other minerals” and will not be discussed in the paper. While the continental samples were dominated by quartz and feldspar, the marine samples were dominated by mica, followed by quartz and feldspar.

In Fig. 8a–c the results of the mineralogical investigation of the eight dust samples chosen are presented. Figure 8a again depicts the average composition of the samples per sampling site (N=4). The minerals zeolite, anhydrite, garnet, sepiolite, fluellite, kaolinite and smectite were only found in the marine samples. Only the continental sample of 15 August–15 September 2014 contained traces of zeolite. While gibbsite, serpentine, calcite and dolomite were detected in the continental dust samples, these minerals were absent in all marine samples. The absence of calcite and gibbsite may have been caused by the pretreatment of the marine sediment trap samples with HCl. Although the concentration of the acid used is fairly low (10 %) and the exposure time of the samples was exactly 1 min, we cannot exclude the possibility that carbonate minerals were dissolved. Therefore, the absence of these minerals in the marine traps will not be discussed further.

In the following, the seasonality in the average mineralogical composition will be outlined for each site as given in Fig. 8b and c. At site Iwik, the winter dust samples were characterized by the occurrence of chlorite, serpentine and rutile, while the summer samples were characterized by the minerals gibbsite and dolomite. At site CBi, the winter dust samples were characterized by the occurrence of the minerals sepiolite, fluellite, kaolinite, smectite, garnet and anhydrite, while the summer samples were characterized by the mineral rutile. Only for the marine trap samples could an annual average chlorite / kaolinite ratio (C / K = 4) be derived owing to the occurrence of kaolinite.

In Figs. 9–12 the results of the 4-day back-trajectory analysis are presented for each sample which has been analysed for mineralogical composition. Four heights (10, 100, 3000 and 4500 m) were chosen to cover both low- (trades) and high-level (SAL) dust transport. A back trajectory was drawn for the day when a dust storm event occurred as depicted on satellite images. Only the low-level back trajectories were plotted for site Iwik because of the correlation of the measured dust characteristics with the low-level wind speed. Moreover, the MWAC samplers were designed to only sample dry deposition, whereas the marine sampling sites collect material settling through the water column, i.e. dust resulting from both dry and wet deposition. The back trajectories at 5500 m can be found in Appendix A.

Figure 9 illustrates a typical late-winter situation. During the sampling interval of each site, at least 2 days with dust storms occurred (Fig. 9c, d). Therefore, two back trajectories were drawn for each height for site CBi and Iwik, respectively. The high-level back trajectories ending at site CBi pass either through the major PSA 2 or point offshore. Both the low-level back trajectories ending at the continental trap site Iwik and at the oceanic trap site CBi point to a dust source within the major PSA 2 (Scheuvens et al., 2013). Some calcite was present in the continental dust sample, but no chlorite or kaolinite was detected. Therefore, the dust source was most likely located in the nearby southwestern Reguibat Shield where sediments are rich in calcite and quartz and depleted in chlorite and kaolinite (Fig. 9a). Dust deposited in the marine traps during the time interval was characterized by the occurrence of chlorite and kaolinite. Thus, the source area of the samples was most likely the chlorite- and kaolinite-rich sediments located near the Boukra phosphate mine in Western Sahara (Fig. 9b).

Figure 10 represents a typical early-winter situation. During the sampling interval of site Iwik, at least three dust storms occurred and at site CBi at least two dust storms occurred (Fig. 10f–h). Each dust storm lasted for several days for which we could model as many as 15 back trajectories for site Iwik and 8 for site CBi for each height. The large number of back trajectories complicated the determination of the likely dust source areas. All back trajectories pass through the major PSA 2 and some point to the PSA 1 and 3 (Scheuvens et al., 2013). One high-level back trajectory ending at site CBi passes through PSA 2 and two through Mauritania and Senegal. However, most of the high-level back trajectories ending at site CBi point offshore. Dust sampled in the marine traps during this sampling interval did not contain any chlorite, while the dust trapped at Iwik did. Chlorite may have been supplied to Iwik from a source area nearby the Senegal–Mauritania Basin (Fig. 10a) or as far as the eastern Taoudeni Basin (Fig. 10b) because of the anomalously high chlorite content of the soils in these areas. The continental sample is further characterized by the occurrence of calcite and the absence of kaolinite, which fits the soils of the source areas chosen (Fig. 10a, b). The marine sample was characterized by the occurrence of zeolite and absence of chlorite. Therefore, zeolite may have been derived from the extrusive volcanic rocks of the northern Taoudeni Basin (Fig. 10c). A further source area might be the southern shoreline of Western Sahara, in which chlorite depleted sediments are situated (Fig. 10d). The presence of the mineral kaolinite in this marine winter sample may be explained by a kaolinite-rich source area lying in the southern Senegal–Mauritania Basin (Fig 10e).

In Fig. 11 a typical early-summer situation is presented. Only one dust storm event was observed during the sampling interval at both sites, which lasted for 1 day (Fig. 11c), resulting in only one back trajectory per site and per height. The low-level back trajectory ending at site CBi runs offshore and the high-level back trajectory passes through the major PSAs 2, 3 and 5. The low-level back trajectory ending at site Iwik passes through the major PSA 2 (Scheuvens et al., 2013). Dust sampled on land at site Iwik was characterized by the absence of chlorite, kaolinite and calcite, which fits the soils of northern Tidra Island (Fig. 11a) making it a really local phenomenon. In contrast, dust sampled offshore at site CBi was characterized by chlorite and by the absence of kaolinite, which fits the chlorite-rich soils in the Mauritanides of Mauritania (Fig. 11b).

In Fig. 12 a typical late-summer situation is illustrated. At least five separate dust events could be identified (Fig. 12f–j), of which three occurred during the sampling interval of site Iwik and two during the sampling interval of site CBi. One of these dust storms occurring during the sampling interval of site CBi lasted for 2 days (7–8 August 2014), while all other dust storms lasted for only 1 day. As a result, three back trajectories could be drawn for each site and each height. The low-level back trajectories ending at site CBi run offshore. The low-level back trajectories ending at site Iwik pass through the major PSA 2. The high-level back trajectories pass through the major PSA 2, 3 and 4 (Scheuvens et al., 2013). Dust deposited in the continental traps was characterized by the presence of calcite and the absence of chlorite and kaolinite. Therefore, the source area of the dust was most likely in Western Sahara, where soils rich in calcite but poor in chlorite and kaolinite are located (Fig. 12a, b). Dust sampled with the oceanic traps during this sampling interval was characterized by the absence of chlorite and kaolinite and by the presence of a high percentage of zeolite (22 %) (Fig. 8c). Therefore, a possible source area may have been extrusive volcanic rocks of the northern Taoudeni Basin (Fig. 12c) and the Fezzan uplift (Fig. 12e). Ferroglaucophane may have been sourced by the Pharusian belt (Fig. 12d).

In Fig. 13a–d the mean wind vectors and speed are presented for the dust storm events chosen. The individual dust source areas that were identified using the back trajectory of the day with the dust storm as shown in Figs. 9–12 are further displayed in Fig. 13a–d. As can be clearly seen in the subfigures, the mean wind velocities were anomalously large in the dust source areas chosen which enabled dust emission. During winter, six-hourly mean wind velocities were larger than 7 ms-1 in the dust source areas chosen (Fig. 13a–b). During summer 2013, six-hourly mean wind velocities were larger than 6 m s-1 in the dust source area chosen (Fig. 13c). During summer 2014 extremely high mean wind velocities were encountered near the study sites and in the dust source area enabling dust emission and transport from a more distant source to site Iwik (Fig. 13d).

In Table 7 an overview of the dust source areas chosen for site Iwik and CBi is given together with the characteristic minerals of the samples that may be used as a tracer for the source area. In the following subsections the identification of the source areas and mineralogical tracers is described in detail.