On the basis of daily Saharan dust samples collected at Sal Island (Cape Verde Archipelagos) and Barbados (Caribbean Sea) over 3 years, this study focuses on the mineralogical signature of the African sources providing dust over the tropical North Atlantic Ocean. First, the sources of the collected dust were localized by using relative clays abundance (illite-to-kaolinite ratio) combined with Meteosat infrared imagery, horizontal visibility, and backward trajectories of dusty air masses. Then, each identified source was linked to a single value of the illite-to-kaolinite ratio. Those results highlight that the clay content of the emitted dust depends directly on both the latitude and the longitude of the source. Dust originating from northwestern sources exhibits illite-to-kaolinite ratios higher than those from Sahelian regions. Likewise, illite-to-kolinite ratio decreases from west to east.
 Mineral aerosols, produced by wind erosion in arid and semiarid regions, contribute to ∼45% of the total atmospheric aerosol load. Their global emission rate can be >1000 Mt yr−1. (For instance, Duce et al.  estimated a range between 1000 and 3000 Mt yr−1.) Because mineral aerosols can be long-range transported, they contribute in a large part to the deep-sea sedimentation [Chester et al., 1979; Rea et al., 1985], provide nutrients to sea surface waters [Duce et al., 1991] and continents [Swap et al., 1992], and during their atmospheric transport affect the radiative budget of Earth's atmosphere by absorbing and/or backscattering solar radiation [Tegen et al., 1996; Li et al., 1996].
 This interaction between mineral dust and climate depends on both dust fluxes and characteristics, mainly size distribution and mineralogy. Conversely, atmospheric dust content is highly sensitive to climatic changes. For instance, dust concentrations in ice core covering the last glacial period recovered in Greenland [Thompson and Mosley-Thompson, 1981], Antarctica [e.g., Petit et al., 1981], and China [Thompson et al., 1989] show that dust fluxes were 5–20 times higher during the Last Glacial Maximum than during interglacial periods. These major increases have been interpreted as a result of drastic aridification and expansion of potential source areas. In turn, recent works have shown that the amplitude of the radiative forcing induced by dust can vary with its mineralogical composition. More precisely, values of the imaginary refractive index in visible and infrared wavelengths depend both on the contribution of each mineral species (silica, alumino-silicates, and carbonates) and on the way they are mixed [Claquin et al., 1998; Sokolik et al., 1998].
 Interaction between climatic changes and dust emissions is mainly assessed by simulations of the atmospheric dust cycle with global circulation models for past and present conditions. However, up to now, most of these simulations [Genthon, 1992; Joussaume, 1993; Mahowald et al., 1999] have failed to reproduce the measured increase in dust concentration for the Last Glacial Maximum. Also, for present day these models do not reproduce the seasonal latitudinal shift of the Saharan dust plume that follows the shift of the Intertropical Convergence Zone (ITCZ) [e.g., Tegen and Fung, 1995]. The reasons invoked to explain these discrepancies between simulations and observations are the difficulties in representing dust source locations and in configuring processes of dust emission.
 However, on the basis of samples collected at Sal Island, Barbados, and Miami during the summer period, Glaccum and Prospero  have shown relative homogeneity in the global mineralogical Saharan dust composition. These authors concluded that the dust source material is well homogenized over a large area of northwest Africa. However, it must be noted that all three dust outbreaks studied by Glaccum and Prospero were originating from the same source region, located roughly near 25°N and 2°E. A recent study [Caquineau et al., 1998] emphasized a large variability in the mineralogical composition of some Saharan aerosols collected at Sal Island (Cape Verde Islands) during the winter season. Such a variability has been linked to three different origin sectors identified by Chiapello et al. , covering potential sources areas from northwestern Africa to Sahelian regions. Those results do not disagree with those of Glaccum and Prospero  but emphasize the strong relationship between mineral dust composition and its source area.
 Thus the present study aims to link the mineralogical characteristics of Saharan dust to their source areas and to estimate the amplitude of variation of their mineralogical signature. For that we developed a methodology based on the combination of relative clay abundance, Meteosat infrared imagery, horizontal visibility, and backward trajectories of dusty air masses. Because of the high spatial and temporal variability of dust emission, added to logistics problems, field campaigns remain difficult to perform within the Saharan region. For these reasons, Saharan dust has been sampled in rather remote regions of the north tropical Atlantic Ocean. Since the major part of Saharan dust is transported westward [Schütz, 1980; D'Almeida, 1986], we have studied dust transported over the tropical North Atlantic Ocean. In order to account for the various spatial and temporal scales involved in the processes of dust emission and transport, two remote areas were selected, Barbados and Sal Island.
2.1. Sampling Sites
 Two principal processes of Saharan dust transport over the tropical North Atlantic Ocean have been clearly identified. In summer, dust transport occurs at high altitudes in the so-called Saharan Air Layer, which typically can reach 5–6 km. This process allows the dust to be transported as far as the Caribbean Sea and controls the seasonal cycle of dust concentrations, which exhibits a maximum between April and October [Prospero, 1996]. By contrast, measurements made at the Cape Verde Islands have shown that maximum dust concentrations occur in winter between November and April. At this time of year, Saharan dust is transported within the trade wind layer, below 1.5–3 km height [Chiapello et al., 1995], and affects only the eastern tropical North Atlantic. Therefore, in order to take into account the seasonal variation of the dust transport, we used samples collected at Barbados (13°10N, 59°30W) during the summer period and at Sal Island (16°45N, 22°57W) during the winter period. The collection and analysis of all these samples are described by Caquineau et al. . Daily concentrations of mineral aerosols recorded on Barbados and Sal Island between December 1991 and December 1994 are given on Figure 1. Because dust inputs at Barbados are quite homogeneous, 20 dust samples representing monthly maximum concentrations and corresponding to 12 dust outbreaks have been studied. By contrast, dust inputs at Sal are characterized by intensive pulses of short duration. We studied 13 of the strongest outbreaks, each one represented by 2–6 samples. The huge dust event that impacted on both Barbados and Sal Island in early April 1994 has been discarded. Indeed, even if this dust outbreak is of great interest for transport study since it occurred simultaneously on both sampling sites [Caquineau et al., 1998], it concerns areas of emission much too large to be usefully interpreted in terms of source region.
 Saharan dust is mainly composed of a mixture of minerals such as quartz, feldspars, and clays, sometimes associated with calcite and/or palygorskite. Mineralogical analysis has been performed according to an analytical procedure specifically developed for our atmospheric dust samples. An original preparation procedure consists in a mineral mass transfer from the initial membrane filter onto a ceramic tile in order to increase the sample mass density. Scans are performed on an X-ray diffractometer fitted with a position-sensitive detector and a collimated incident beam. The semiquantitative treatment is carried out according to the method ofChung , using reference intensity factors determined from field samples. Analytical procedure and semiquantitative treatment are fully described by Caquineau et al. .
 Relative percentages in mass for quartz, illite, and kaolinite are reported on Figure 2. As described in section 1 [Glaccum and Prospero, 1980], a relative homogeneity in the mineralogical distribution of these three major constituents characterizes the samples collected at Barbados, while those collected at Sal Island are characterized by a larger variability.
 We focused our attention on clay minerals, their ratio of relative abundance having been shown to vary with dust origin [Chester al., 1972; Kiefert and McTainsh, 1995; Caquineau et al., 1998]. For instance, the ratio between relative abundance of illite and kaolinite (I/K ratio) was shown to be a relevant fingerprint of the regional origin of Saharan dust. Indeed, this ratio is not affected by the fractionation processes that occur during both the emission and the long-range transport of the dust [Caquineau et al., 1998]. The use of this ratio as a mineralogical tracer, measured in the dust collected at Sal Island, allowed us to confirm the three main sectors of origin previously defined by Chiapello et al. . Regarding the I/K ratio measured in Barbados samples (Figure 2), except for a few values around 0.5 corresponding to a geographical sector already identified (south and central Sahara), most of the values are spread around 1. This can either indicate a mixture of dust originating from the three sectors identified or the contribution of an extra sector.
 Finally, assuming that a high variability of the I/K ratio probably reflects the contribution of more than one source, dust events that do not exhibit a constant mineralogical signature all along the event have been discarded. However, if stable I/K ratios can be used as indicators of dust origins, they do not provide any information on the geographic location of the sources. Thus mineralogical data were associated with complementary analytical tools providing information on the origin of the aerosols analyzed in that way.
2.3. Satellite Observation
 For clear conditions during daytime the presence of atmospheric mineral dust over the arid areas of North Africa results in a global decrease of the longwave radiance outgoing to space. This effect, described in early studies [Shenk and Curran, 1974; Legrand et al., 1983, 1985; Oliva et al., 1983], is the basis of an algorithm designed for the remote sensing of desert dust, using the thermal infrared channel (10.5–12.5 μm) of Meteosat satellites (pixel resolution varies between 160 km by 160 km and 225 km by 225 km from Sahelian to Moroccan regions). The method is described by King et al.  as the infrared contrast method. The corresponding algorithm is described by Legrand et al. . The main steps are recalled hereinafter. From a time series of Meteosat IR images at midday a composite image is constructed by selecting for each pixel the clearest observation (i.e., the maximum radiance). The resulting image, called the reference image, approximates an image for clear and dust-free conditions. A period of 15 days is found to be a good compromise [Legrand et al., 1994] for both a satisfactory elimination of dust and water clouds and a limitation of the impact of the seasonal effects. Each original image is then subtracted from the reference image, providing a difference image that contains only the atmospheric elements, dust and water clouds. Water clouds are identified using an algorithm derived from the spatial coherence method of Coakley and Bretherton . At last, pixels are classified as cloudy or clear. The former are masked, while the latter represent the Infrared Difference Dust Index, or IDDI, expressed as a difference of radiometric counts or as a brightness temperature decrease. On all the daily satellite images presented herein, increasing values of IDDI are represented by colors, from black for dustless pixels to red for high dust amounts. Red patches represent values of IDDI larger than 38 counts, which corresponds to values of visible optical depth >1.5 [Legrand et al., 2001]. The masks of water clouds are displayed in gray.
2.4. Horizontal Visibility
 Horizontal visibility has been used to check the validity of the IDDI as a dust indicator [Legrand et al., 1994; Chomette et al., 1999]. We have compared data of horizontal visibility with 77 IDDI images for the 3-year period selected for our study. For each of the 32 meteorological stations selected (between 12°N and 35°N and between 17°W and 34°E), the IDDI values are computed for the 3 × 3 pixel arrays centered at the stations, provided that all the pixels are clear. In Figure 3 the data are classified into seven classes of horizontal visibility measured at 1200 UTC. As predicted, we observe a strong correlation between the two data sets, IDDI increasing as the visibility decreases. It reaches a value of ∼10 counts for a visibility of 10 km (this correspondence is identical to the result reported by Chomette et al. , derived from visibility data of 39 stations in 1984). For horizontal visibility higher than 10 km, the correlation is not so good. However, this disagreement is not of interest in this study since dust is considered to be absent in this case. The observed relation confirms that the highest IDDI values are associated with the smallest visibility values, which are related to the presence of dust emission.
2.5. Back Air Mass Trajectories
 Backward air mass trajectories allow us to reconstruct the atmospheric pathway of dust between the continent and the sampling site and therefore to help identify the source region that has emitted the collected dust at a given date. So they act as a link between the mineralogical signature of the transported dust and their source region.
 Three-dimensional isosigma back trajectories with a horizontal resolution of 2.5° × 2.5° have been calculated by using the TM2z version [Ramonet and Monfray, 1996] of a three-dimensional atmospheric transport model initially developed by Heimann . The dispersion of an initial tracer mass is calculated for a maximum of 5 days by using the inverted wind fields from the European Centre for Medium-Range Weather Forecasting (ECMWF) as input data. Twice a day (at 0000 and 1200 UTC), the position of the center of gravity of the tracer mass is updated. In this study, the point of highest probability of the dispersion plume is considered as the air mass position. Ignoring the deviation between the calculated ECMWF wind fields and the actual unknown wind fields, the uncertainty on the air mass position is given by the width of the retrodispersion plume, which increases with the number of days. After 5 days it is ∼250 km.
 According to the altitude of the dust transport in winter (upper limit between 1.5 and 3 km [Chiapello et al., 1995]), trajectories ending between 0.4 and 1.2 km height above the station of Sal Island were calculated. Each of these trajectories was then associated with the corresponding daily samples, allowing us to reconstruct the atmospheric pathway of the dust for the 5 days preceding its sampling date. Because the model used is not able to provide backward trajectories for more than 5 days, air mass trajectories could not be calculated for outbreaks of dust transported over Barbados. Indeed, satellite imagery (principally, that of GOES and advanced very high resolution radiometry (AVHRR)) shows that it takes ∼1 week for dust outbreaks to cross the tropical North Atlantic Ocean from the African coast to the Caribbean [Ott et al., 1991]. Satellite images were then interpreted without the help of air mass trajectories regarding these dust outbreaks.
3. Case Studies
 For all the selected dust outbreaks our strategy consisted in looking at temporal series of daily satellite images preceding the date of maximum concentration of dust recorded at the sampling stations. The length of those temporal series is variable not only according to the sampling site (3–10 days) but also from one case to another. In the case of large cloud covering, satellite images are replaced by fields of horizontal visibility when available. The visualization of the areas of emission at a given date is not always sufficient to identify the source of the dust effectively transported up to the sampling site, in particular when more than one source is involved. In this case, back air mass trajectory observations are a useful additional tool providing information when available.
3.1. Outbreaks Observed at Sal Island During Winter
3.1.1. Northwestern Sahara origin
 A dust outbreak was recorded at Sal Island in 1994, between February 19 and 21. A maximum mineral dust concentration of 190 μg m−3 was measured on February 20. Before the 19 and after the 21, dust concentrations were <45 μg m−3. The corresponding daily samples are associated with the northwestern Saharan sector as previously defined by Chiapello et al. , and their mineralogical composition is characterized by values of I/K ratio ranging from 2.2 to 2.6, indicating the large dominance of illite. On February 15 the infrared satellite image shows no dust over northern Sahara. Infrared satellite images on February 16 and 17 are presented on Figure 4a. Backward trajectories ending at Sal Island between February 19 and 21 have been reported day by day on the satellite images in order to visualize the position of the air mass at the date of the satellite image. On these trajectories the colored circle indicates the position of the air mass at the time when the image was created. A source is considered as identified when coincidence between areas of high IDDI (orange or red color) and air mass position is observed.
 On February 16, dust was emitted in the northwestern Sahara, approximately in southern Morocco (orange color). On the 17, while this dust cloud moves along the coast toward the south, a second source becomes active in Algeria (red color). Finally, on February 18 (not shown), dust coming from southern Morocco has disappeared, whereas dust coming from Algeria covers almost all the central Sahara. This temporal series suggests that two distinct source regions located in southern Morocco and Algeria could be responsible for this dust event. However, the weak variability of the mineralogical composition of dust during the 3 days (February 19–21) is coherent with the contribution of a single source. By reporting the air mass trajectories on all the images, we note that the air mass arriving at Sal Island on February 20 (the day with the maximum dust concentration at Sal) is coming directly from southern Morocco. Conversely, the air mass arriving at Sal Island on February 21 came close to the Algerian dust source on February 17, but the dust concentration of February 21 is decreasing at Sal Island. Dust collected at Sal Island on February 21 could partly originate from the Algerian source, but it would make only a minor contribution to the mineralogical signature. We therefore assume that the major source of this event is southern Morocco, characterized by an I/K ratio ranging between 2.2 and 2.6 (with a mean of 2.3).
 The dust collected at Sal Island between February 8 and 13, 1994, exhibits a mean I/K ratio of 1.6. For the first time, satellite images (Figure 4b) show that on February 4, two regions are simultaneously active in the northwestern Sahara. Initially in contact, the two dust clouds clearly separate on February 5. Reporting the air mass trajectories ending at Sal Island on February 8 and 9, the days of the maximum dust concentration, clearly allows us to identify the origin of dust. The two air masses have an oceanic path until February 4, and they both pass over the westernmost source on February 5. An I/K ratio of 1.6 can then be associated with this source, located in northern Mauritania. This source differs from the example previously detailed (Morocco with an I/K ∼2.3) by both its location and its mineralogical signature.
3.1.2. Sahelian origin
 During the whole sampling period, only two dust events have shown an I/K ratio close to 0.1 (January 3–6, 1992, and January 14–18, 1992). Satellite images corresponding to the first one are shown on Figure 4c. The two air mass trajectories ending at Sal Island on January 5 and 6, 1992, at 1200, are reported on the images of December 31, 1991 and January 2, 1992, respectively; they exhibit the same zonal westward pathway. On December 31, 1991, three sources located at relatively low latitudes (noted a, b, and c from west to east on Figure 4c) display a strong activity (red color). On January 2, 1992, the satellite image shows that dust coming from the source c is most probably transported toward the Gulf of Guinea and therefore cannot reach Sal Island. The two other source areas (noted as a and b on Figure 4c) are masked by clouds. Consequently, horizontal visibility data were examined for ground stations situated nearby the source areas a and b. On Figure 5 one can observe that the air masses reaching Sal on January 5 and 6, 1992, originate from a region situated between Agadez (Niger) and Gao (Mali) where horizontal visibility was ∼300 m between December 31, 1991, and January 2, 1992. This region corresponds to the source noted b on the satellite image of December 31. Consequently, the dust observed at Sal Island between January 3 and 6 was attributed to a source region covering northwestern Niger and northeastern Mali and characterized by a mean I/K ratio of 0.1.
3.1.3. Central and southern Sahara origin
 At Sal Island, January 1993 was marked by successive dust outbreaks, two of them showing homogenous mineralogical characteristics, with I/K ratios of ∼0.4. Satellite images associated with these outbreaks show that a same and unique region, located in South Algeria, has displayed a strong activity during the whole month. However, whatever the dust outbreak, the air mass trajectories never reached this source within 5 days, which represents the limit for the air trajectory model used. As an example, the case of the dust event recorded at Sal Island between January 14 and 16, 1993, is reported on Figure 4d. Strong emissions from southern Algeria (source area a) and Sahelian regions (source area b) are obvious on satellite images of January 9 and 10, 1993, but potential sources can also be hidden by clouds in the northwestern Sahara. However, the mineralogical characteristics for this event (I/K = 0.4) are neither consistent with a northwestern Saharan origin nor with the fingerprint of the Sahelian origin (I/K = 0.1).
 Taking into account the dust I/K value and its low variability, the systematic activity of this Algerian source (noted a) during the month, and the similarity of the pathway of all the corresponding air masses, we have considered that the dust collected at Sal Island during January 1993 mainly originated from the south Algerian source (noted a), which is characterized by an average I/K ratio of 0.4.
3.1.4. Other dust outbreaks
 The dust outbreak observed between March 9 and 12, 1992, is an example of a mixture of two dust clouds originating from two different sources: in northern Sahara (northern Algeria) and in southern Sahelian regions. The mineralogical signature of the dust collected at Sal Island between March 9 and 12, with an I/K ratio of 1.1, integrates the characteristics of each of these two source regions. Considering that the southern and Sahelian sources are characterized by I/K ratios close to 0.4 and 0.1, respectively, the only way to observe a final I/K ratio equal to 1.1 is to affect an I/K ratio distinctly higher than 1 to the north Algerian source.
 Finally, on the 10 winter outbreaks observed at Sal Island and exhibiting a constant I/K ratio, three of them could not be connected to source regions because of cloud presence over North Africa. Unfortunately, fields of horizontal visibility could not be used because of a lack of data for these three events. On the other hand, a total of seven dust events collected at Sal Island during the winter season could be used to the identify five emitting sources. These sources are spread along a NW-SE axis, from northern Algeria (north of the Grand Western Sandsea, between Ghardaia and Bechar), Morocco (between the Anti-Atlas mountains and the Drâa Hamada), and north Mauritania (around the Sebkha Ghallamane) to south Algeria (Tanezrouft, east of Bordj-Mokhtar) and Niger (southeast of Adrar des Iforas mountains and west of Agadez).
3.2. Outbreaks Observed at Barbados During Summer
 Regarding the summer outbreaks recorded at Barbados (Figure 1), seven cases out of the 12 studied were discarded for two major reasons: (1) the presence of numerous clouds over North Africa due to the northern position of the ITCZ at this time of year and (2) the difficulty of locating durable potential source areas. Because each dust event was only represented by one or two samples, it was not possible to check the variability of I/K ratios and consequently to check the uniqueness of the dust origin. On the other hand, calcite generally associated with palygorskite has been detected in several samples collected at Barbados and can be used as a qualitative signature of sources, different from those already identified.
3.2.1. Egyptian origin
 We present a detailed analysis of an outbreak recorded at Barbados at the end of summer, between October 4 and 6, 1992. On October 5 the maximum dust concentration was 67 μg m−3, and the mineralogical characteristics of the corresponding sample are an I/K value of 0.7, with the presence of calcite and palygorskite. There are no active sources before September 25. Satellite images recorded between September 25 and 28, 1992 (Figure 6a), show that two regions can be involved during this period. Between September 25 and 27, dust is only present over the northwestern part of the Sahara. From September 28 an intense emission of dust occurs in Egypt. The following images (not shown) indicate a westward movement of the latter dust cloud. There is no other active source in the following days.
 The contribution of the northwestern area is discarded because the I/K ratio measured in the dust was not consistent with the mineralogical signature of the western sources previously identified from winter outbreaks at Sal Island. We show in section 3.1 that dust originating from Morocco and Mauritania is characterized by I/K ratios of 2.3 and 1.6, respectively. Consequently, it can be assumed that dust observed on Barbados on October 5, 1992, originated from the source located in Egypt, which is therefore characterized by an I/K ratio close to 0.7 along with presence of calcite and palygorskite. The location of this source means a transport duration to Barbados of 7 days that seems rather short. However, such a summer transport, taking place at high altitude, corresponds to an air mass velocity of 15–20 m s−1, which is not unrealistic.
3.2.2. Other outbreaks
 According to the same procedure, four other sources were identified for this period of the year. They are situated in central Algeria (area centered on In Salah), along the Libyan-Tunisian border, in central Libya (south of Sirte desert), and in south Algeria (Tanezrouft).
18.104.22.168. South Algerian and Libyan-Tunisian origin
 During July 1994, continuous Saharan dust input was observed at Barbados, leading to a background dust concentration close to 20 μg m−3. A maximum concentration of 50 μg m−3 was measured on July 12, with a dust I/K ratio of 1.1 and presence of palygorskite. Satellite observations between June 30 and July 7 show that several areas are affected by dust emission. For example, on satellite images of July 1 and 2 (Figure 6b), different sources emitted dust between the Libyan-Tunisian border and the Mauritanian coast and in south Algeria. However, the presence of palygorskite can be considered as a characteristic of the easternmost sources (Libyan-Tunisian border) since this mineral has never been observed in dust originating from the sources of the northwest Sahara and south Algeria.
22.214.171.124. Libyan and central Algerian origin
 The dust outbreak observed between April 10 and 16, 1993 (maximum dust concentration of 80 μg m−3 on April 12), shows how extensive the source region can be in summertime. The satellite image recorded on April 1 shows that an extensive dust uplift occurs simultaneously from central Algeria to Libya and Egypt (Figure 6c). On April 2 the Algerian dust cloud is transported westward while dust emission continues in the Libyan and Egyptian areas. There is no other dust emission during the following days.
 Consequently, no mineralogical signature could be attributed to these sources because they were involved in the production of dust but were never active separately. Additional observations are required to assess their mineralogical signature.
 From the dust outbreaks studied, we report (Figure 7) the mineralogical signature (I/K ratio) of emitted dust on a map of identified Saharan source regions.
4.1. Location of Dust Sources
 The location of the identified sources seems consistent with previous works. From a degradation soil map, personal observations, and indications provided by local observers, D'Almeida  identified four principal dust source areas reported on Figure 7. One extends within western Sahara. The two westernmost sources we have identified in this work are included in this large source area. Likewise, the source situated north of the Adrar des Iforas mountains (southern Algeria) might correspond to a second major source identified by D'Almeida. On the basis of meteorological data associated to mineral dust fallout in western Mediterranean, Bergametti et al.  identified two sources, both situated north of the thirtieth parallel (Figure 7). One of them located between eastern Algeria and northwestern Libya is consistent with the source we identified in southeastern Tunisia and northwestern Libya.
 In this study, we only localized sources that have produced westward transported dust during 3 years (1992–1994), but other sources may exist. For example, Bertrand et al.  have reported the existence of a huge source located south of the Aïr and Tibesti Mountains and lying up to Lake Chad (Figure 7). We also observed strong dust emissions from this region, especially during winter (December and January). However, since infrared satellite images showed that the emitted dust was essentially transported southwestward toward the Gulf of Guinea, following the direction of the Harmattan wind, this source area was not reported in this work.
 The sources we identified in the eastern part of the Sahara do not correspond to previously identified source regions. Results of D'Almeida  indicate two sources in these areas: The first one is located in southern Libya, northeast Niger (between east of Aïr and north of the Tibesti Mountains), and the other one is located in southern Egypt, northern Sudan. Both sources are located more southerly than the two sources we identified in central Libya and north Egypt. According to back air mass trajectories reported by Chester et al. , central Libya could be a potential source area for dust collected in the Tyrrhenian Sea. Likewise, Ganor and Foner  situated the possible sources of most of the dust storms reaching Israel in two regions: one covering Chad and Libya and another covering Egypt and the Libyan and Negev Deserts. Most of the available information concerning the sources situated in eastern Sahara was derived from studies of northeasterly transport of dust. Except for our own results, cases of westward transport of dust from these areas had never been reported. This needs to be documented more thoroughly since we only investigated a limited number of cases.
4.2. Mineralogical Signature of Saharan Dust Sources
Figure 7 shows that the mineralogical signature of emitted dust from western sources strongly depends on the latitudinal position of the source. Indeed, we observe a gradient from the northern sources (I/K = 2.3) to the southern ones (I/K = 0.1). This latitudinal evolution is consistent with the mineralogical characteristics of Atlantic sediments [Windom, 1975] and of dust collected over the Atlantic Ocean [Chester et al., 1972]. Schütz and Sebert  have observed in Saharan soils an increase of relative amounts of kaolinite toward the south. This feature was also observed in dust collected along a path from 35°N to 19°N [Paquet et al., 1984]. Likewise, illite dominates the clay fraction of the south Moroccan soils [Coudé-Gaussen and Rognon, 1993], whereas kaolinite is the major constituent of aerosols collected in Niger [Drees et al., 1993; Sabre, 1997]. Kiefert et al.  mentioned a K/I ratio of 2.6 (equivalent to an I/K of 0.4) for local dust collected in Mali, which is in good agreement with the I/K ratios we found for our southern sources. These findings can be explained by the influence of lateritic soils that become more and more frequent in the Sahelian area [Paquet et al., 1984]. Additionally, the north Algerian source was only estimated with an I/K ratio higher than 1 (see section 3.1.4). This estimate is consistent with the mineralogical composition of aerosols collected in northern Algeria (south of the Atlas Mountains) and exhibiting an I/K ratio of 2 [Gomes, 1990].
 This study also characterizes the mineralogical signature of a source located in Egypt. The low associated I/K ratio (0.7) of this northern source is surprising given the global latitudinal dependency of the I/K tracer. However, data from Ganor and Foner  show that dust originating from Libya and Egypt and reaching Israel are characterized by mean concentrations of illite and kaolinite of 15 and 30%, respectively, which is consistent with the I/K ratio of 0.7 found for the Egyptian source.
 A second conclusion can be drawn from the map. Our results indicate that the measured I/K ratio also apparently depends on the longitude of the source region. More exactly, the kaolinite content of the dust increases from west to east. In order to check the consistency of this longitudinal evolution, and because of a lack of field measurements in eastern Sahara, we compare the mineralogical composition of Saharan dust collected at various parts of the Mediterranean.
 By analyzing 11.5 years of daily monitoring of dust transport retrieved from archives of Meteosat shortwave (visible light spectrometer) and infrared (IR) images, Moulin et al.  have shown that the major source regions for the Mediterranean area were Morocco and northern Algeria for the western basin; Tunisia and northern Algeria for the western central basin; Libyan Desert, northern Algeria, and southern Tunisia for the central eastern basin; and the Egyptian desert for the far eastern Mediterranean. These results imply that, at a regional scale, the western Mediterranean is more affected by dust coming from western Saharan sources, and, conversely, the eastern Mediterranean is more affected by eastern sources. As a consequence, the longitudinal variation of the mineralogical composition of the Mediterranean dust should be related to the longitudinal change of the source mineralogy.
 The mineralogical composition of Saharan dust collected in the Mediterranean area existing in the literature is plotted on Figure 8 as illite and kaolinite concentrations. All the data correspond to atmospheric dust samples (dry deposition) except those of Avila et al.  which were obtained from rain samples containing Saharan dust (the so-called “red rains”). As expected, Figure 8 clearly shows differences in the Saharan dust composition from west to east, characterized by a decrease of the illite content along with an increase of the kaolinite concentration. These findings are in agreement with our observations and indirectly confirm the longitudinal dependency of the mineralogical signature (I/K ratio) of the Saharan sources. More systematic investigations of the mineralogical composition of Saharan dust transported toward the Mediterranean Sea, in relation to its origin, would provide useful data to improve the mineralogical signature of the eastern sources and to complete our map. More generally, the multidisciplinary approach developed in this study can be considered as relevant enough to determine the mineralogical signature of the Saharan sources, without the need for sampling in situ the source areas.
 The purpose of this study was to link the variation of the mineralogical composition of westward transported Saharan dust to its sources. Combining complementary tools such as images of an infrared satellite dust index, the IDDI, horizontal visibility, and backward trajectories allowed us to associate a value of the illite-to-kaolinite ratio with the identified sources. We have shown that the mineralogical signature varied as a function of both source latitude and longitude. The I/K ratio progressively decreases from 2.3 to 0.1 from northwest Sahara to central Sahel and from west to east. Those trends are consistent with previous studies and, more generally, with the distribution of clay species at the regional scale of the Sahara. However, such an approach has proved successful to improving both the Saharan source location and its mineralogical signature, without the need for in situ sampling inside the source areas.
 Highlighting this direct relationship between a source region and the mineralogical characteristics of emitted dust is of primary interest in various scientific fields. For instance, it could facilitate the interpretation of the paleoclimatic record based on the dust contained in ocean sediments and ice cores. Being able to determine the origin of this mineral dust, according to its mineralogical composition, would greatly improve our knowledge of past and present-day atmospheric circulation. Moreover, by improving location of the source regions, our results can be used to test the capacity of the dust emission scheme to reproduce their location. Finally, mineralogical characteristics of dust directly affects its optical properties through the refractive index, a key parameter for modeling its radiative impact.
 We are particularly grateful to J.M. Prospero, who provided filters and data relative to the Barbados samples. We especially acknowledge G. Cautenet for his help in providing us the meteorological data. We are indebted to B. Chatenet, who made possible the measurements on Sal Island, to P. Bousquet and I. Chiapello for their help in the compilation of back air mass trajectories, and to G. Gabalda and T. Pilorge for their programming support. We also acknowledge the two reviewers for their valuable comments. This work was supported by the “Programme Environnement du CNRS” in the framework of the action “Erosion Eolienne en Régions Arides et Semi-arides.”