Journal of Geophysical Research: Atmospheres

Uplift of Saharan dust south of the intertropical discontinuity

Authors


Abstract

[1] In situ observations from a flight made during the Geostationary Earth Radiation Budget Intercomparison of Longwave and Shortwave Radiation (GERBILS) field campaign (June 2007) show significant dust uplift into the monsoon flow immediately south of the intertropical discontinuity in the western Sahara. Dust loadings were highest in the moist monsoon air and the observations are consistent with dust uplift by the nocturnal monsoon winds. There is some evidence that cold pools within the monsoon flow contributed to the dust uplift: regions of elevated dust, water vapor, and ozone within the monsoon air are consistent with precipitation cooling and moistening air from upper levels and the resultant dusty cold pools propagating northward. However, only southward propagating cold pool outflows could be observed in satellite imagery. Using European Centre for Medium-Range Weather Forecasts analyses and satellite data, it is shown that the asymmetry in the seasonal dust cycle is closely related to the downdraft convective available potential energy (DCAPE) from convective storms. There is both more dust and more DCAPE during monsoon onset than during retreat. The larger DCAPE values during monsoon onset, as well as the stronger nocturnal monsoon flow and the stronger heat trough circulation, are expected to contribute to the higher dust loadings at this time. Both the monsoon flow and cold pool outflows within it result in dust uplift in the western Sahara during the monsoon onset, which is when the maximum dust uplift occurs. For dust modeling, this shows the importance of accurately modeling not only the monsoon flow itself but also deep convection and cold pools.

1. Introduction

[2] The importance of desert dust in weather and climate processes is well recognized [Prospero and Lamb, 2003; Jones et al., 2004; Haywood et al., 2005; Mahowald et al., 2005; Tompkins et al., 2005; Evan et al., 2006; Pérez et al., 2006; Field et al., 2006], and the Sahara desert is the world's largest dust source [Tanaka and Chiba, 2006]. Dust uplift is a complex function of wind, soil type, soil wetness, and vegetation and is normally parametrized as depending on the low-level wind speed cubed, with a threshold wind speed [e.g., Marticorena et al., 1997]. This nonlinear dependence on wind speed can make small-scale fluctuations in wind speed very significant. Such gustiness occurs on scales smaller than those resolved by climate models. Many studies have highlighted the role of boundary layer convection in generating these gusts [e.g., Cakmur et al., 2004; Engelstaedter and Washington, 2007a; Marsham et al., 2007], but there are other important sources, such as cold pool outflows from convective storms, “haboobs” [Sutton, 1925], and other density currents [Bou Karam et al., 2008].

[3] Within the Sahara spatial variations in soil types, vegetation, and climate result in some particularly intense dust sources, for example, the Bodele depression and certain locations in the western Sahara [Prospero et al., 2002]. For these sources in the western Sahara Engelstaedter and Washington [2007a] showed that the annual cycle in dust uplift appears to be related to a convergence zone north of the intertropical convergence zone. Engelstaedter and Washington [2007a] hypothesized that this dust uplift is due to strong boundary layer convection in this convergence zone but noted that the southward passage of the convergence is not associated with such uplift, and in arid Saharan zones this cannot be due to changes in vegetation. Bou Karam et al. [2008] identified the northern convergence zone as the intertropical discontinuity (ITD) and described airborne lidar observations of dust uplift into the head of the turbulent monsoon flow, which at the ITD was behaving as an intrusive density current. Bou Karam et al. [2008] suggest that this process may make a significant contribution to the seasonal cycle observed by Engelstaedter and Washington [2007a].

[4] There is a pronounced diurnal cycle in the monsoon flow in West Africa [Parker et al., 2005]. During the day boundary layer convection inhibits the southwesterly monsoon winds, which as a result are most intense at night. For the dust source regions of the western Sahara, Engelstaedter and Washington [2007b] showed a better correlation between European Centre for Medium-Range Weather Forecasts (ECMWF) “gustiness” and dust than between ECMWF winds at 1200 UTC and dust. Knippertz [2008] suggests that this is because the gustiness parameter better captures the morning maximum in low-level winds, as the momentum of the nocturnal flow is mixed downward toward the surface [Parker et al., 2005], since the ECMWF gustiness parameter is the maximum wind from 0900 to 1200 UTC derived from the grid-scale 10 m wind speed, vertical wind shear, and vertical stability. Grid-scale winds at 1200 UTC cannot be expected to capture this morning maximum in wind speed. Using data from May and June 2006, Knippertz [2008] discusses how this process is most significant on the dry side of the ITD, where the stronger nocturnal cooling leads to a stronger nocturnal inversion and, consequently, greater near-surface wind speeds as the low-level jet is mixed downward during the morning. Knippertz [2008] also suggests that the asymmetry in dustiness between monsoon onset and retreat, which was observed by Engelstaedter and Washington [2007a], may be related to the stronger heat trough circulation during monsoon onset because of the greater contrast between land and ocean at this time.

[5] During the monsoon, deep convection typically leads to rainfall reaching the surface in all regions approximately 200 km south of the ITD [Hamilton and Archbold, 1945], with mesoscale systems most common at night [Nesbitt and Zipser, 2003]. Between the ITD and this region, rainfall often evaporates before it reaches the ground but can still generate haboobs [Hamilton and Archbold, 1945]. Haboobs from storms south of the ITD can also propagate northward and merge with the ITD itself [e.g., Flamant et al., 2007] or be located just to the south of the ITD [e.g., Bou Karam et al., 2008]. Recently, Williams [2008] has highlighted both the contribution of dust uplift by haboobs to the seasonal cycle in dustiness in the Sahel and the possible masking by convectively generated cirrus of much of the dust uplifted by Sahelian haboobs from the seasonal cycle observed by Engelstaedter and Washington [2007a].

[6] Miller et al. [2008] give a useful review of haboobs, which have recently begun to receive increased scientific attention. These dust storms, generated by the downdrafts from convective storms, are typically 1–2 km deep and accompanied by a sharp decrease in temperature and an increase or decrease in relative humidity. Away from the generating storm, however, dust uplifted by cold pool outflows can become mixed with the environmental air and precede the sharp gradient in thermodynamic variables and strong winds. Miller et al. [2008] use idealized modeling to show that, especially when synoptic-scale wind speeds are low, haboobs can be responsible for a significant fraction of the dust uplift that occurs from the southern Arabian Peninsula. Until recently, there have been few published observations of these features in Saharan and sub-Saharan Africa, despite their common occurrence and their importance for the uplift of Sahelian dust during monsoon onset, which is the main season for dust uplift [Sterk, 2003]. Sutton [1925, 1931] and Lawson [1971] describe observations from Sudan. Knippertz et al. [2007] describe observations from the Atlas Mountains in Morocco, and Knippertz [2008] shows satellite imagery of haboobs in the Sahel and western Sahara. Flamant et al. [2007] and Bou Karam et al. [2008] describe airborne lidar observations from the Sahel, and Cuesta et al. [2008] list eight haboob events affecting Tamanrasset (in Algeria) during May-September 2006. Williams et al. [2008] note that virtually every cold pool outflow from moist convection in Niamey (in the Nigerian Sahel) during the wet season of 2006 was a substantial source of dust uplift. Visual observations made by the authors confirm that even cold pools from precipitating cumulus congestus can result in visible amounts of dust uplift in the Sahel during monsoon onset.

[7] The Geostationary Earth Radiation Budget Intercomparison of Longwave and Shortwave Radiation (GERBILS) field campaign took place in June 2007 and made six flights between Niamey in Niger (13°N, 2°E) and Nouakchott in Mauretania (18°N, 16°W). Measurements were made close to the western Saharan source regions described by Prospero et al. [2002] and were referred to as WA1 and WA2 by Engelstaedter and Washington [2007a] (approximately 18°–20°N and 2°–10°W). During GERBILS several cases of cold pool outflows leading to dust uplift were noted in satellite imagery, but only one flight provided data from a region where it appeared that this may have occurred (B301 on 27 June 2007). Data from this flight are described in this paper (section 3.1) together with data from the next day's flight (B302), which essentially reversed the track of B301. The model and data used in this paper are described in section 2. The observations from West Africa presented in section 3 are consistent with the observations by Bou Karam et al. [2008] of significant uplift occurring in the moist turbulent monsoon air south of the ITD, including uplift within cold pools. Using satellite imagery and ECMWF reanalyses, section 4 relates the seasonal cycle in dust observed in satellite data to the seasonal cycle in the energy available to downdrafts from deep moist convection. The conclusions are summarized in section 5.

2. Models and Data Used

[8] Version 3.19 of the Consortium for Small-scale Modeling (COSMO) model [Doms and Schättler, 2002] was run in forecast mode during the GERBILS field campaign. COSMO is a nonhydrostatic regional model, and equations for fully compressible flow in a moist atmosphere are formulated in rotated geographical coordinates and terrain-following height coordinates. The horizontal grid spacing was 0.0625° (approximately 7 km), and 35 levels were used in the vertical grid (grid spacing of approximately 50 m at the lowest level). Deep convection was parametrized using the Tiedtke convection scheme [Tiedtke, 1989]. The model was initialized with ECMWF operational analyses, and the boundaries were forced with three hourly ECMWF forecast fields (the COSMO model domain is shown in Figure 1b). Results from two simulations are used in this paper. These were initialized at 0000 and 1200 UTC on 27 June. Where results are shown at the time of initialization these are referred to as the COSMO analysis since these are essentially ECMWF analysis fields interpolated onto the COSMO grid. The COSMO forecast, as well as being at a higher resolution than the ECMWF analysis, allowed a comparison between model fields and aircraft observations at the time of observation (ECMWF analyses were only available at 0000, 0600, 1200, and 1800 UTC).

Figure 1.

Model fields with the east to west flight track of flight B301 superimposed. (a) ECMWF analysis of dew point temperature at 1000 hPa at 1800 UTC on 26 June 2007 (filled) and 0600 UTC on 27 June 2007 (solid lines). The contour interval is 5 K in both cases. The 285 and 290 K lines are drawn for both days (dashed lines for 26 June). (b) WVMRs of 950 hPa and winds from the COSMO analysis at 1200 UTC on 27 June 2007. The high-level and low-level legs of B301 are shown by thick and thin white lines. The aircraft position at 1200 UTC ± 7.5 min is shown by the black line. (c) Contours of 10% total cloud cover on 27 June 2007 from the COSMO analysis at 1200 UTC (solid lines), COSMO forecast at 1200 UTC (dashed lines), and the maximum cloud cover from the COSMO forecast between 0000 and 1200 UTC (dotted lines).

[9] The Facility for Airborne Atmospheric Measurements BAE 146 was instrumented to measure the standard meteorological and chemical parameters as well as aerosols. Temperatures were measured using Rosemount platinum resistance immersion thermometers and water vapor by a Lyman-Alpha absorption hygrometer. Data from upward- and downward-pointing shortwave broadband (0.3–3 μm) radiometers were used to calculate surface albedo. The aerosol scattering coefficient (i.e., the scattering cross-sectional area of the aerosols per unit volume of air) measured by a TSI 3563 integrating nephelometer is used to quantify atmospheric dust loadings. Observations were made at red (700 nm), green (550 nm), and blue (450 nm) wavelengths, but only the blue wavelength observations are used as only relative dust loadings are required in this study. All low-level flight legs were flown at constant pressure levels (approximately 600 m above the ground for flight B301 and 400 m for B302).

[10] Meteosat imagery was used to detect dust and clouds (the pixel size is approximately 3 km in West Africa). The images shown in Figure 2 are constructed from the 8.7, 10.8, and 12 μm channels, with the difference between the 12 and 10.8 μm channels displayed in red, the difference between the 10.8 and 8.7 μm channels displayed in green, and the 10.8 μm channel shown in blue. As a result water vapor appears blue and the thermal contrast between the warm land surface and cooler elevated dust means that the dust appears pink [Ackerman, 1997].

Figure 2.

False color Meteosat imagery from 27 June 2007 with the B301 flight track superimposed. The key to the imagery is also shown. Plots are shown for (a) 0030, (b) 0400, and (c) 1200 UTC. Systems A, B, B2, and C are referred to in the text.

[11] In section 3.2, in addition to ECMWF analyses, various satellite products are used. The International Satellite Cloud Climatology Project (ISCCP) D2 monthly mean data set [Rossow et al., 1996] is used to show the seasonal cycle in deep convection and cirrus on a 0.5° grid. Global Total Ozone Mapping Spectrometer (TOMS) aerosol index (AI) products, available on a 1.25° longitude by 1.0° latitude grid, are used to show dust [Herman et al., 1997]. Aerosol index is a measure of the difference between the ultraviolet radiance observed and that expected from a purely gaseous atmosphere. Aerosols that are absorbing in the UV (e.g., dust and smoke) give positive AIs [Prospero et al., 2002]. TOMS AI is sensitive to the height of the boundary layer, but in the western Sahara Engelstaedter and Washington [2007a] noted similar seasonal cycles in infrared dust difference index produced from Meteosat and TOMS AI. In addition to TOMS AI, retrievals of aerosol optical depths (AODs) from the Multiangle Imaging Spectroradiometer (MISR) are used to detect dust (level 3 products on a 0.5° grid [Kahn et al., 2005]). Retrievals of leaf area index (LAI) from the Moderate Resolution Imaging Spectroradiometer (MODIS) [Yang et al., 2006] and volumetric soil moisture from the Advanced Microwave Scanning Radiometer-EOS (AMSR-E) [Njoku, 2004] are also used.

3. Observations From GERBILS of Dust Uplift Close to the ITD on 27 June 2007

[12] ECMWF analyses show that between 1800 UTC on 26 June 2007 and 0600 UTC on 27 June there was a decrease in the geopotential at 1000 hPa in West Africa, and the southwesterly nocturnal monsoon flow led to a northward shift in the ITD (often defined by a dew point of 287 or 288 K [e.g., Eldridge, 1957; Buckle, 1996]) over the region (Figure 1a). This was more pronounced in the northward bulge in the ITD situated over northern Mali (approximately 0°–6°W). On 27 June, in order to start the low-level leg of flight B301, the aircraft descended northwestward, to a point just south of the northern edge of the moist monsoon flow (Figure 1b). The false color satellite images (Figure 2) show that this was close to where deep convection had occurred during the preceding night.

[13] Figure 2a shows that at 0030 UTC three mesoscale convective systems (MCSs) had formed close to the ITD (labeled as A, B, and C) and close to the locations later observed by flight B301. A smaller storm (B2, approximately 200 km across at 0030 UTC) can be seen on the northern edge of storm B. The COSMO forecast and analysis gave deep convection in approximately the correct locations, although there were errors in the mesoscale structures (Figures 1c and 2). Precipitation in COSMO only reached the surface south of 15°N, whereas data from the Tropical Rainfall Measuring Mission showed precipitation at up to 16.3°N (not shown). No cold pool structures were detectable north of 16°N in COSMO. However, especially in West Africa where the observational network is limited, errors in mesoscale cloud cover, precipitation, and cold pool structures are to be expected, especially from a model in which deep convection is not explicitly resolved [Diongue et al., 2002].

3.1. Aircraft Observations and Comparison With the COSMO Model

[14] Data from the northwestward descent to the westward low-level leg of flight B301 and from the low-level leg itself are shown in Figures 3 and 4with the corresponding fields from the COSMO model. The low-level leg started at approximately 1115 UTC, from a point that was located under the clouds remaining from system B2 at approximately 0400 UTC (Figure 2b).

Figure 3.

Data from the descent to the low-level leg of B301 (27 June 2007; 16.9°N, 2.3°W to 17.9°N, 3.4°W; 1054–1114 UTC) and the corresponding COSMO forecast fields (heights are above mean sea level, and horizontal dotted lines show the land surface height at the northern end of the descent). Solid lines show aircraft data. Dash-dotted lines and dotted lines show vertical profiles from COSMO at locations at the start and end of the descent, and dashed lines show data from COSMO along the slanted descent made by the aircraft. The positions of the monsoon air, SAL, and free troposphere are also shown (A–E refer to near-neutral layers within the SAL, and “residual layer” indicates a possible residual layer from the previous day's monsoon). (a) Potential temperature (thicker red lines) and water vapor mixing ratio (thinner purple lines). (b) Observed aerosol scattering coefficient (σ, blue line) and vertical wind (small black crosses). (c) Wind speed (black lines) and direction (red lines).

Figure 4.

Data from the low-level leg of B301 (27 June 2007). Solid lines show aircraft data, and the dashed lines show forecast COSMO fields interpolated to the corresponding locations and times. Dotted lines show COSMO analysis fields at 1200 UTC. (a) Potential temperature (θ, red) and water vapor mixing ratio (purple). (b) Virtual potential temperature (θv, red) and wind speed (green). The maximum COSMO forecast wind speeds at 10 m are shown from 0000 to 0900 UTC (long-dashed line) and from 0000 to 1200 UTC (dash-dotted line). (c) Westerly wind (u wind, black) and southerly wind (v wind, red). (d) Aerosol scattering coefficient (σ, blue) and ozone concentration (red). (e) Height of orography (black) and aircraft altitude (blue). Vertical dotted lines show the boundaries of the regions used in Figure 6. These are labeled Saharan air, ITD, suspected cold pool, monsoon, and SCP (suspected cold pool).

[15] The observations in Figure 3 show three main layers: the free troposphere (above 5500 m), the Saharan Air Layer (SAL) (between 2500 and 5500 m), and below this the monsoon, which is separated into two layers by a lid at approximately 1500 m (all heights in this paper are given above sea level unless otherwise stated). The SAL shows multiple near-neutral layers, each with a different dust and moisture content (labeled A–E in Figure 3). This is a common feature of the SAL, as is the local maximum in dust and the water vapor mixing ratio (WVMR) observed at the top of the layer [e.g., Flamant et al., 2007].

[16] Since observations in Figure 3 are from a slanting aircraft descent, it is possible that the two moist layers below 2500 m may have been separated horizontally rather than vertically. However, the upper moist layer (1500–2500 m) was not as turbulent as the layer below (as shown by the vertical winds in Figure 3b). This suggests that only this lowest layer was an active boundary layer that was fully coupled to the surface. In addition, only the lowest layer is well mixed and this lowest layer was 3.5 g kg−1 moister than the layer above. This is consistent with a residual layer from the previous day where monsoonal and SAL air have mixed. The scattering coefficients in the lower layer reached 600 × 10−6 m−1, indicating very high concentrations of dust. This is approximately 3 times higher than that observed in the layer above.

[17] Winds were approximately westerly in these two monsoon layers and easterly in the SAL and free troposphere above (Figure 3c), with the main change in wind direction occurring over a depth of less than 500 m (2100–2500 m, at the top of the upper monsoon layer). Wind speeds were approximately 15 m s−1 in the lower monsoon layer and 5 m s−1 in the lower part of the SAL. The main change in wind speed occurred across the upper monsoon layer between 1500 and 2200 m, i.e., lower than the main change in wind direction.

[18] COSMO only shows this observed two-layered structure in the monsoon flow in a profile from the southeast end of the slanted aircraft profile (dash-dotted line in Figure 3a). In addition it shows the upper layer as being continuously stable rather than initially well mixed and stable above. Figure 3 shows that although the observed WVMRs in the lowest level (below 1500 m) are consistent with COSMO (dashed line), those in the layer above (1500–2500 m) are only consistent with the southern COSMO profile from the start of the descent (dash-dotted line), while WVMRs at approximately 3000 m are closer to the later northern profile from COSMO (dotted line). The fit between COSMO and the data could therefore not be significantly improved by artificially offsetting the modeled monsoon northward or southward. COSMO shows reasonable agreement with the observed wind speeds and directions in the lower monsoon layer and in the lower part of the SAL. The change in wind speed in COSMO occurs approximately 200 m too low down, but the magnitude of the shear is in reasonable agreement with the observations. The change in wind direction in COSMO occurs over too great a depth, i.e., 1500–2500 m rather than at the top of the upper monsoon layer between 2100 and 2500 m.

[19] In summary, the aircraft descended northwestward through the SAL to reach the monsoon boundary layer, which by this time (1114 UTC) was well mixed by dry convection, with a dew point of approximately 289 K (not shown). Between this boundary layer and the SAL there was a layer approximately 1 km deep, which was possibly the residual layer from the previous day's monsoon flow after it had mixed with the SAL during the day. The monsoon boundary layer contained much higher dust loadings than the air above. The COSMO model captured the broad features of this profile but not the fine structure of the layerings observed.

[20] The low-level leg westward at 18°N showed a moisture front at 6.2°W (Figure 4a). To the east of this front dew points were between 288 and 291 K, while to the west dew points were between 283 and 286 K (not shown). This moisture front at 6.2°W therefore fulfils the definitions of the ITD used by Eldridge [1957] and Buckle [1996] (i.e., dew point thresholds of 287 or 288 K, respectively). This moisture front is therefore referred to as the ITD. There was very little change in the winds or buoyancy at this front, however (Figures 4b and 4c), and it is possible that a definition of ITD based on low-level convergence may have resulted in an alternative location (as discussed by Hastenrath [1985]). West of the ITD the air was approximately 4 g kg−1 drier and less dusty. East of the ITD three regions of distinct WVMRs and dust contents were observed (approximately 3.4°–3.7°W, 3.7°–5.0°W, and 5.0°–6.2°W, shown by vertical dotted lines in Figure 4). The COSMO forecast captured the location of the ITD at 6.2°W but not its sharpness. It also gave WVMRs that were approximately 1 g kg−1 too high west of the ITD and did not show the cold air between 5° and 6°W and east of 4°W. The COSMO analysis gave a sharper ITD, but this was approximately 1° too far west and still not as sharp as observed. WVMRs east of 6.2°W were 1–3 g kg−1 too low in the analysis, and the analysis also did not show the cold air between 5° and 6°W and east of 4°W. The COSMO forecast captured the observed wind speed variations reasonably well but overestimated the wind speeds between 4.3° and 6°W by approximately 2 m s−1.

3.2. Discussion of Dust Uplift Mechanisms

[21] By 1030 UTC uplifted dust was clearly visible in the false color Meteosat imagery in regions close to the flight track of B301 (not shown). Although west of 6.2°W there are positive correlations between the observed dust loadings and both the observed and modeled wind speeds (Pearson correlation coefficient is 0.8, not shown), no such relationship is seen in the moist air to the east, where most of the dust was observed (Figure 4). Instead, in this moist air dust loadings correlate with WVMRs and anticorrelate with potential temperatures. Similarly, the dust loadings do not correlate with the maximum 10 m wind speeds from the COSMO model, either from the previous night or at the time of the flight (Figures 4b and 4d). The COSMO model captured the observed variations in boundary layer winds at 1200 UTC reasonably accurately (Figure 4b) and the position of the ITD, if not its sharpness. We therefore deduce that the observed variations in dust loadings were either from variations in wind speed on scales unresolved by the COSMO simulations (approximately less than 40 km with the 7 km grid spacing used) or from resolvable processes at times earlier than the observations, which COSMO fails to represent. It is hypothesized that the dust uplift was due to the moist monsoon air acting as an intrusive density current overnight [Bou Karam et al., 2008], turbulent downdrafts and cold pool outflows from the convective storms to the south, or a combination of both mechanisms.

[22] To detect dust, the thermal imagery shown in Figure 2 relies on the temperature contrast between airborne dust and the land surface and so is biased toward dust with temperatures significantly different from the land surface. It is therefore particularly biased to elevated dust during the day. Either the nocturnal monsoon flow or cold pools may have resulted in dust uplift during the night, but the dust may only have become detectable in the satellite imagery once it was mixed upward by boundary layer convection (this effect is discussed by Chaboureau et al. [2007]). Either the monsoon flow alone or embedded cold pool outflows could lead to the largest dust loadings being observed in the coldest moistest air. The weak contrast in virtual potential temperature (θv) and wind speed at 6.2° and 3.8°W (Figure 4b) shows that the coldest wettest air masses were no longer acting as active density currents at the time of flight B301. This is expected from the moisture front of the ITD during the day but is also similar to the aged cold pools discussed by Tompkins [2001] in simulations of tropical convection over the ocean.

[23] The dust, the storms, and dew points from ECMWF analyses are shown in Figure 5. This shows that the dust was in the northward “bulge” in the ITD, where the monsoon air had moved northward during the night (Figure 1a) and close to storms A and B2. Using a combination of visible and infrared Meteosat imagery, arcs of low-level clouds were observed propagating southwestward away from the convective systems A, B, and C (shown in Figure 5). These were presumably “arcus” clouds on the leading edges of cold pool outflows from the storms (they are just visible as pink arcs south of storms A and C in Figure 2c). These features were only detectable to the south of the generating storms, where there was more moisture for cloud formation. They propagated southwestward, against the prevailing winds which were approximately southwesterly overnight and westerly during the day. (The arc of clouds immediately south of the aircraft position at 1200 UTC shown in Figure 2c is at middle and high levels, and no dust features were detectable in the visible imagery there.)

Figure 5.

The locations of the main features observed on the night of 26 June and the morning of 27 June 2007. Labeled contours show 285 K dew point temperatures from ECMWF analyses at 1000 hPa 1800 UTC on 26 June (dashed contour) and 0600 UTC on 27 June (solid contour). The storms (A, B, B2, and C) are shown at 2300 (solid black lines), 0100 (dashed black lines), and 1200 UTC (solid gray lines). Storm A was dissipating at 1200 UTC, and only a letter is used to show its location (B2 had already dissipated). Arcus clouds at the edge of cold pool outflows are shown at 0800 (dotted black lines), 1100 (long-dashed gray lines), and 1200 UTC (dash-dotted gray lines). The main regions of dust uplift detectable in satellite imagery at 1200 UTC (Figure 2c) are indicated by the hatched gray area. The aircraft track is shown, with the nephelometer data shown on the low-level leg (see Figure 4).

[24] If cold pools propagated northward as well as southward from storms A, B, and C, then the locations of the observed dust are consistent with the hypothesis that cold pool outflows from these storms contributed to the dust uplift (Figure 5). However, no sharply defined storm outflows were detected in satellite imagery in these regions of observed dust uplift. It is unsurprising that arcus clouds were not observed because of the dry environment north of these storms. Dusty cold pool outflows can often be seen in visible or infrared satellite imagery [e.g., Knippertz et al., 2007]. However, in the south, although arcus clouds were observed and uplift of Sahelian dust is expected [Sterk, 2003], cold pools were only detectable because of the arcus clouds rather than as colder and brighter regions in infrared and satellite imagery. To the north, dusty outflows could not be detected in the visible or infrared imagery, but obscuration by cirrus made this difficult. As already noted, it can also be difficult to detect dust using infrared imagery until such a time as it is mixed upward by convection. For storm B2 visible imagery was also only of use after 0700 UTC, 5.5 h after the dissipation of the storm, when the leading edge of any cold pool may have become quite diffuse and difficult to detect. We conclude that although northward propagating cold pools from storms A and B2 were not observed in the satellite imagery, it is possible for air from cold pool outflows from storms A and B2 to have reached the location of the low-level flight track of B301 and for this process to have been undetectable in the imagery available.

[25] The dusty lowest layer observed on the descent into the dusty region closest to storm B2 (Figures 3 and 5) is consistent with both the hypothesis that dust was uplifted by the monsoon flow and the hypothesis that a cold pool outflow was responsible. Similarly, the regions of colder air between 5° and 6.2°W and east of 3.8°W, with sharp fronts, are consistent with aged downdrafts from storms but may have been from variations in the monsoon flow not captured by COSMO. The cold wet air certainly does not appear to be related to soil moisture anomalies generated by rainfall since satellite retrievals from AMSR-E detected no surface wetting there and soil moisture would also be expected to lead to decreased dust loadings in moist areas rather than the increases observed [Marticorena et al., 1997].

3.2.1. Analysis of Ozone Data

[26] Figure 4 shows elevated ozone concentrations in the cold wet and dusty air between 5° and 6°W and east of 3.7°W. These locations are consistent with dust uplift from hypothesized cold pools propagating northward from storms A and B2. However, as discussed in section 3.2, cold pools were only observed propagating southward (against the nocturnal southwesterly winds). Cold pool outflows are expected to bring ozone-rich air down into the boundary layer from the SAL above. To analyze this more quantitatively, the low-level leg of flight B301 was split into sections: Saharan air (9°–6.3°W), ITD air (6.3°–6.14°W), suspected cold pools (6.14°–5°W and 3.7°–3.4°W), and monsoon air (5°–3.7°W). These sections are shown in Figure 4 (separated by vertical dotted lines).

[27] Ozone is usually anticorrelated with water vapor since the reaction between the two depletes ozone. Figure 6a shows that for the data set as a whole this is generally true, with the drier Saharan air containing more ozone than the moister monsoon air. However, the suspected cold pools have higher ozone, and in some cases higher WVMRs, than the rest of the monsoon air. Air from the ITD is consistent with a mixture of monsoon or cold pool air and Saharan air, as expected.

Figure 6.

Ozone, WVMR, and aerosol scattering data from the low-level flight of B301 (shown in Figure 4, 3.4°–9°W). Data from Saharan air (9°–6.3°W) are shown in red, data from ITD air (6.3°–6.14°W) are shown in orange, data from suspected cold pools (6.14°–5°W and 3.7°–3.4°W) are shown in purple and blue, and data from monsoon air (5°–3.7°W) are shown in green. Plots show (a) ozone and WVMRs and (b) ozone and aerosol scattering, with a straight line fit for the Saharan and monsoon data shown (solid line). Two-standard error uncertainties were calculated for the gradient and intercept of this line. Using these uncertainties, the dashed lines show the lines with (1) maximum gradient and minimum intercept and (2) minimum gradient and maximum intercept.

[28] Figure 6b shows that for the Saharan, ITD, and monsoon air there is a linear relationship between dust and ozone (There are perhaps two separate linear relationships, one for the Saharan air and one for the monsoon air, with the ITD air being a mixture of these two. Using two such linear fits, which are only significantly different using a 1 σ uncertainty, does not affect our analysis). Such relationships have been observed before [e.g., Bonasoni et al., 2004]. They may be related to heterogeneous reactions of ozone with dust, but there are significant uncertainties in these processes [Mogili et al., 2006]. This uncertainty is not, however, important for our analysis here.

[29] Figure 6b shows that air from the suspected cold pools has increased ozone and dust contents compared to the monsoon air. In order to allow a straightforward calculation of the significances of the differences between the suspected cold pools and the other regions, the data in Figure 6b were rotated so that the best fit straight line shown in Figure 6b was horizontal. The new y values of the data were then approximately normally distributed. Ignoring the spatial autocorrelations in the data, this showed that the dust and ozone contents of the air from both the suspected cold pool regions were significantly different from the air from the other regions (differing by more than 10 standard errors). The ozone concentrations in the suspected cold pools are similar to the Saharan air. These observations are consistent with the hypothesis that storms A and B2 have cooled and moistened ozone-rich air from the SAL by evaporation, and the resultant downdrafts have brought this air down into the boundary layer, where the resultant turbulence has resulted in dust uplift. An alternative hypothesis is that the coldest and wettest regions of the monsoon air form the most active density currents overnight and that these have elevated ozone for some other reason. This is not expected, however, since the reaction between ozone and water vapor depletes ozone (consistent with Figure 6a).

3.2.2. Summary of Possible Dust Uplift Mechanisms

[30] In summary, the results show dust uplifted into the monsoon air immediately south of the ITD. Dust loadings in the monsoon air were not related to observed winds at midday or nocturnal winds from the COSMO forecast. This is consistent with dust uplift occurring through turbulent processes or through processes that occurred before the time of observation that could be resolved by COSMO but were not well represented by the model. The evidence presented is consistent with both the hypothesis that dust was uplifted by the monsoon flow and the hypothesis that a combination of the monsoon flow and cold pool outflows were responsible (although if only the monsoon flow was responsible, the elevated ozone concentrations in the moister dustier regions shown in Figure 6 is unexplained).

[31] The statistically different ozone and dust contents of the colder dustier regions is consistent with the hypothesis that these regions were affected by outflows from convective storms. The locations of these dustier regions is consistent with cold pools propagating northward from the observed storms, although only cold pools propagating southwestward against the winds were observed (Figure 5). For storm B2 the fact that a cold pool was not observed propagating northward in visible satellite imagery is unsurprising since visible imagery was only available 5.5 h after B2 dissipated. Before this time it is possible that a northward propagating cold pool was not detected in infrared imagery because of cirrus cover. If a cold pool did propagate northward from storm A, it must have been obscured by cirrus until its leading edge was too diffuse to identify. Although the details of the contribution to dust uplift by storm downdrafts and cold pool outflows are uncertain in this case, it is known that this process occurs immediately to the south of the ITD [Sterk, 2003; Flamant et al., 2007; Bou Karam et al., 2008]. The seasonal variations expected from uplift by cold pools are investigated in section 4.

3.3. Effect of the Observed Dust Uplift on Albedo

[32] Flight B302 reversed the low-level leg of B301 on the next day (28 June 2007). Dust loadings were much lower for B302 than for B301 (Figure 7), and the aircraft altitude was similar (200 m lower). Figure 7 shows that the differences in albedo observed between B301 and B302 are closely related to the difference in the optical thicknesses of the dust estimated to be below the aircraft in each case (these optical thicknesses (τ) were calculated using the assumption that the dust was uniformly vertically distributed within the boundary layer below the aircraft and neglecting absorption by aerosols, so that τ is the aircraft altitude above ground level (agl) multiplied by the nephelometer scattering coefficient). Therefore, we can conclude that the high dust loadings observed in the moist air east of 6.2°W were responsible for the observed increase in the albedo there, although any small differences in the aircraft track between B301 and B302 are expected to contribute to the observed variability in the albedo difference. From the relatively clear flight B302 to the dusty flight B301, Figure 6 shows that the albedo observed from a height of 600 m above the ground increased by 0.033 (from approximately 0.47 to 0.50, or 7%). This would have decreased surface sensible heat fluxes by decreasing the downwelling solar radiation at the surface, especially as Figure 3 shows that the main dust layer from B301 was approximately 1100 m deep.

Figure 7.

The difference in albedo observed from the aircraft between flights B301 and B302 (black line). The estimated difference in the optical thickness of the dust (τ) between the two flights is shown in red (see text for details). The nephelometer data are shown in blue (solid line for B301 and dashed line for B302).

4. Seasonal Cycle in Dust Uplift by Cold Pools South of the ITD

[33] This section moves on from the case study presented in section 3 to consider the seasonal cycle in the contribution of cold pool outflows south of the ITD to dust uplift using data from a period of approximately 3 years. Figures 8a8f show the seasonal cycle of dustiness using TOMS AI at 6°W, the approximate longitude of the dust source regions WA1 and WA2 discussed by Engelstaedter and Washington [2007a]. Fields from ECMWF analyses (at 0000, 0600, 1200, and 1800 UTC), ISCCP cloud fields, AMSR-E soil moisture, and MODIS LAI fields are superimposed on this TOMS AI data.

Figure 8.

Satellite observations of aerosol at 6°W (colored) with other fields contoured. (a) ECMWF DCAPE for CAPE >100 J kg−1 (contours at 500 (dashed line) and 1000 J kg−1 (solid line)). (b) ECMWF CAPE from 925 hPa (contours at 100 (dashed line) and 500 J kg−1 (solid line)). (c) Daytime deep convective cloud amount from ISCCP (2% contour spacing). (d) Daytime cirrus cloud amount from ISCCP (5% contour spacing). (e) Leaf area index from MODIS (contours at 0.3, 0.5, 1, 2, and 3). (f) AMSR-E volumetric soil moisture times 1000 (contours at 80, 100, and 120 m3 m−3). (g) As Figure 8a but for MISR AOD, not TOMS AI. Figures 8a–8f show TOMS aerosol index (TOMS Earth Probe in 2004; TOMS Ozone Monitoring Instrument from 2005 onward), and Figure 8g shows aerosol optical depth from MISR.

[34] Figure 8a shows TOMS AI and downdraft convective available potential energy (DCAPE) for locations where convective available potential energy (CAPE) is greater than 100 J kg−1 and energy is available for moist convection (Emanuel [1994] defines both CAPE and DCAPE, and these were both calculated from ECMWF analyses, with CAPE for the 925 hPa level and DCAPE mass averaged up to 20 hPa). Figure 8a shows that the seasonal cycle in dustiness observed using TOMS is closely related to the cycle in the energy available to downdrafts in storms (i.e., DCAPE). DCAPE is higher during monsoon onset than retreat since the free troposphere is drier at this time (not shown). A dry midtroposphere consistently leads to more intense cold pools, as inferred by Barnes and Sieckman [1984]. This asymmetry in DCAPE between monsoon onset and retreat is expected to contribute to the observed asymmetry in dust uplift between these times.

[35] The asymmetry in dust uplift cannot be explained by the amount of energy available to deep convection (CAPE) or the area of deep convective clouds observed, which are similar during monsoon onset and retreat (Figures 8b and 8c; note that although the peak in cloudiness at 20°N in December and January is real, there is very little CAPE and any convection is much less intense than the convection associated with the monsoon). It is also interesting to note that Figure 8c shows that the first maximum in TOMS AI in 2004 (during March and April) coincides with a peak in deep convective cloud, and Figure 8a shows that TOMS AI values are higher during the retreat phase of years with higher values of DCAPE. However, given the limited lengt of these data sets these effects may be coincidences. Figure 8d shows that differences in cirrus cloud cover (which may mask some dust from TOMS, especially in the Sahel, as discussed by Williams [2008]) cannot explain the differences in TOMS AI between monsoon onset and retreat. Changes in vegetation coverage and near-surface soil moisture also affect dust uplift. The detectable changes in these do not extend beyond 17°N, however (Figures 8e and 8f), and are expected to be important for Sahelian, rather than Saharan, dust sources.

[36] Although MISR retrievals of AODs are poorest for dusty atmospheres over bright land surfaces, it has been found to be useful in such regions [Kahn et al., 2005]. Figure 8g shows that the asymmetry between monsoon onset and retreat is less clear in MISR AODs than in TOMS AI but is detectable and is clearest in the years with the clearest asymmetry in DCAPE (2005 and 2006). MISR shows more dust further north than TOMS AI, with the maximum at 30°N, rather than in the monsoon flow (probably uplifted by strong low-level northerly winds associated with flow splitting around the northwest flanks of the Hoggar, as discussed by Flamant et al. [2007]). In addition to this maximum, MISR shows a maximum in dustiness at the northern limit of the CAPE associated with the monsoon (reaching 25°–27°N), which is distinct from the main maximum further north. These differences between the trends shown by TOMS AI and MISR AOD may be related to the differences in the cloud masking used for TOMS AI and MISR AOD products, the difficulties in retrieving AODs from MISR over bright desert surfaces [Kahn et al., 2005], and the possible bias of TOMS AI against dust at low levels (discussed by Engelstaedter and Washington [2007a]), which in particular may explain why TOMS misses the dust in the north.

[37] In summary, using data from a period of approximately 3 years, Figure 8 shows that during this period there is more energy available to downdrafts from storms during monsoon onset than retreat, which is when the most dust is observed. In addition, greater nocturnal monsoon wind speeds during monsoon onset are expected to lead to greater dust uplift by the monsoon flow at this time [Sultan et al., 2007; Bou Karam et al., 2008], and a stronger heat trough circulation is expected to lead to more dust uplift, especially to the north of the ITD [Knippertz, 2008]. Furthermore, the upward transport of dust as it moves toward the axis of the heat trough is also expected to enhance the TOMS AI values since TOMS AI is biased toward elevated dust [Knippertz, 2008].

[38] Cakmur et al. [2004] show that in the Sahara as a whole, and for their model, the effects of boundary layer convection on dust uplift dominated those from subgrid turbulence in cold pools but do also show that the maximum contribution of the turbulence from cold pools is in June. Moist convection, and the uplift from it, was parametrized by Cakmur et al. [2004], and this is expected to lead to significant errors in the timing and location of convection in West Africa [e.g., Yang and Slingo, 2001], as well as in the uplift itself. Although our results and others [e.g., Sterk, 2003; Flamant et al., 2007; Bou Karam et al., 2008; Knippertz, 2008] show that the turbulent monsoon flow and cold pool outflows uplift significant quantities of dust in the western Sahara, the role of such processes in the Sahara as a whole is not well quantified; if this quantification is to be improved, the monsoon flow, deep moist convection, cold pools, and the resultant dust uplift must be accurately represented in models.

5. Conclusions

[39] At around midday on 27 June 2007 aircraft observations from GERBILS showed high dust loadings in a region of the western Sahara south of the ITD (defined by any dew point threshold between 286 and 288 K). The observations show high dust concentrations in the moist monsoon air and lower concentrations in the drier Saharan air. Wind speeds were approximately 14 m s−1 at the aircraft altitude of approximately 600 m above the ground but only correlated with the observed dust loadings in the drier Saharan air. In the monsoon air the highest dust loadings were found in the coldest moistest regions. Dust uplift in the moist monsoon air appears to be a result of either (1) processes that occurred before the time of observation (around midday) that were resolvable by a model with a 7 km grid spacing but were not well represented or (2) turbulent processes that occurred before the time of observation. These are, as far as the authors are aware, the first published in situ observations of dust uplifted at the ITD in the Sahel or the Sahara.

[40] The ITD advanced northward on the nights of 26 and 27 June, when several active MCSs were present to the south. The observations are consistent with both the hypothesis that observed dust uplift occurred as a result of the monsoon flow and the hypothesis that both the monsoon flow and northward propagating cold pools were responsible for the dust uplift. In satellite imagery cold pools were only observed propagating southwestward (where arcus clouds formed) against the modeled boundary layer winds. However, in the drier environment to the north, it is possible that cold pools did exist and did not generate arcus clouds but could not be detected in the satellite imagery. Elevated ozone concentrations were observed in the coldest, moistest, and dustiest air, which are consistent with dust uplift by cold pool outflows; if only the monsoon flow led to the dust uplift in these regions, then the higher ozone concentrations remain unexplained. If cold pool outflows contributed to the observed dust uplift, then the in situ observations presented show that downdrafts from storms can contribute to dust uplift in this region, even when this is not immediately clear from satellite imagery. The maximum in deep convection is at night, and cold pool outflows are at low levels. This can make detection of the uplifted dust using visible and infrared imagery difficult until the dust is mixed upward during the day.

[41] By midday the colder regions observed were barely acting as active density currents, presumably as during the morning they had been warmed by the large surface sensible heat fluxes. Compared with observations from a much less dusty flight the next day, the observed dust loadings led to an observed increase of albedo from approximately 0.47 to 0.50, or 7% (when viewed from approximately 600 m agl within the dusty monsoon air). This is expected to have reduced surface fluxes, while the airborne dust is expected to have absorbed solar radiation. The effects of these processes on dusty density currents should be investigated.

[42] Uplift in cold pools south of the ITD is not unusual [e.g., Sterk, 2003; Flamant et al., 2007; Bou Karam et al., 2008; Knippertz, 2008; Williams et al., 2008]. Using satellite retrievals and ECMWF model fields, we show that the seasonal cycle of dustiness observed in satellite data from the western Sahara is closely related to the cycle of moist convection and in particular the energy available to downdrafts in convective storms. There is more dust uplift during monsoon onset, when DCAPE is higher, than during monsoon retreat. This asymmetry in DCAPE over the seasonal cycle is expected to contribute to the asymmetry observed in the seasonal cycle in dust uplift observed in the western Sahara by Engelstaedter and Washington [2007a]. Nocturnal monsoon winds are also expected to be larger during monsoon onset [Sultan et al., 2007], and Knippertz [2008] suggests that the stronger heat trough circulation during monsoon onset is also expected to increase dust uplift at this time, especially on the dry side of the ITD. Furthermore, the elevation of the dust in the heat trough may contribute to the seasonal cycle in TOMS AI [Knippertz, 2008]. Using satellite retrievals, we have shown that although seasonal changes in vegetation and soil moisture are significant in the Sahel and perhaps the southern Sahara, they cannot explain the seasonal cycle of dust observed further north.

[43] We conclude that cold pool outflows are one of the processes resulting in dust uplift in this region during monsoon onset, when the maximum dust uplift occurs. The relative contributions to dust uplift in this region from cold pool outflows and the monsoon flow, as well as other processes only briefly referred to in this paper, such as momentum mixed downward from low-level jets, orographically forced flows, and synoptically driven wind events, should be further investigated. In order to model dust uplift in West Africa models must accurately represent the low-level jets, monsoon flow, deep convection, cold pool outflows, and the resultant dust uplift. The boundary layer dynamics of such models must then represent the transport of this dust within the boundary layer and the SAL.

Acknowledgments

[44] This project was funded by the Natural Environment Research Council (NERC) (NER/O/S/2002/00971 NE/B505538/1) and funding obtained through AMMA-EU. On the basis of a French initiative, AMMA was built by an international scientific group and is currently funded by a large number of agencies, especially from the European Community's Sixth Framework Research Programme. Detailed information on scientific coordination and funding is available on the AMMA International Web site http://www.amma-international.org). We would like to thank Ben Johnson (Met Office, UK) for helping to organize the GERBILS field campaign; the staff of FAAM, whose hard work allowed the GERBILS field campaign to run so smoothly; Gary Robinson (NERC Environmental Systems Science Centre, University of Reading) for providing Meteosat imagery from the RADAGAST project; Mat Evans (University of Leeds) for discussions about the ozone chemistry; the UK Met Office for providing access to the ECMWF data; Conny Schwierz (University of Leeds, UK) for facilitating access to this ECMWF data; and Sarah Jones (Universität Karlsruhe and Forschungszentrum Karlsruhe, Germany) for enabling use of the COSMO model. Finally, we would like to thank the reviewers, in particular Peter Knippertz and Cyrille Flamant, whose comments enabled significant improvements to be made, both to content and clarity.

Ancillary