Corresponding author: J. H. Marsham, University of Leeds, National Centre for Atmospheric Science, Leeds, UK. (firstname.lastname@example.org)
 We describe observations from the Fennec supersite at Bordj Badji Mokhtar (BBM) made during the June 2011 Fennec Intensive Observation Period. These are the first detailed in situ observations of meteorology and dust from the central Sahara, close to the center of the Saharan heat low and the summertime dust maximum. Historically, a shortage of such Saharan observations has created problems for evaluating processes, models, and remote sensing. There was a monsoon influence at BBM before 8 June and after 12 June, with dry Harmattan winds in between. A split boundary layer, generated by ventilation from the Atlantic, persisted during the drier phase. Extensive cold pools (haboobs) and microburst-type events were regularly observed. Moisture reached BBM at night from the monsoon and the embedded haboobs. As well as the regularly occurring nocturnal low-level jet (LLJ), a Saharan upper boundary layer (650 hPa) jet was observed, where winds feel drag from dry convection in the afternoon. This jet is linked to the diurnal cycles of moisture and cloud. Most dust was observed in the cloudier monsoon-affected periods, and covarying dust and cloud amounts explain most of the variations in shortwave radiation that control the surface sensible flux. Dustiness is related to a standard parameterization of uplift using 10 m winds (“uplift potential”), and this is used to estimate uplift. Around 50% of uplift is nocturnal. Around 30% is from the LLJ, and 50% is from haboobs, which are mainly nocturnal. This demonstrates, for the first time from observations, the key role of haboobs, which are problematic for models.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 The atmosphere above the Sahara is a key component of the Earth's climate system. The strong solar irradiance combined with large-scale subsidence over the North African landmass leads to an extremely arid environment, with some of the highest near-surface temperatures on Earth, and the deepest boundary layer [Cuesta et al., 2009]. In summer, this leads to the formation of the Saharan heat low (SHL), which is a key component of the West African monsoon system [Parker et al., 2005; Lavaysse et al., 2009] that brings the main annual precipitation to the millions who rely on rain-fed agriculture in the Sahel. The Sahara is also the largest source of mineral dust to the Earth's atmosphere [Prospero et al., 2002; 2005Washington et al., 2003]. This dust interacts directly with both infrared and solar radiation, and it has been shown that prognostic dust can improve weather prediction [Tompkins et al., 2005]. Saharan dust particles also act as ice nuclei for clouds, provide essential nutrients to the biosphere, and darken snow and ice surfaces when deposited, and dust events are a hazard for transport and human health [Carslaw et al., 2010].
 The Fennec project aimed to produce, for the first time, a comprehensive meteorological data set from the Sahara, since despite the Sahara's importance to Earth's climate, it has one of the sparsest networks of routine meteorological measurements of any landmass on Earth (white pluses, Figure 1), with the majority of these measurements being made on the periphery of the desert. There have also been very few field campaign observations: ground-based Saharan observations from AMMA (African Monsoon Multidisciplinary Analysis) were based in Tamanrasset, Algeria, which is a mountainous area unrepresentative of the central desert [Cuesta et al., 2008]; other projects such as GERBILS (Geostationary Earth Radiation Budget Intercomparison of Longwave and Shortwave Radiation [Haywood, 2011]) and SAMUM (Saharan Mineral Dust Experiment [Ansmann et al., 2011]) have been based around the margins of the Sahara.
 Fennec's extended observation period began in May 2011 with the deployment of near-surface meteorological instruments, and the intensive observation period (IOP) took place in June 2011. During the IOP, airborne observations were made from a base in Fuertventura (Figure 1) [Ryder et al., 2013]. The ground-based field campaign is a cornerstone of Fennec and has three components: high-resolution, multiparameter observations at two main supersites, SS1 at Bordj Badji Mokhtar (BBM), Southern Algeria, and SS2 at Zouerate, central Mauritania (Figure 1; M. C. Todd et al., Meteorological and dust aerosol conditions over the Western Saharan region observed at 1 Fennec supersite-2 during the Intensive Observation Period, submitted to Journal of Geophysical Research, June 2011), and less comprehensive observations of surface meteorology from a network of remote automatic weather stations distributed across the SHL region [Hobby et al., 2013]. The SSs are located as close as possible to the climatological centers of the summertime SHL and aerosol maximum as logistics allowed and were heavily instrumented during the IOP. Instruments at SS1 and SS2 were operated by the Office National de Météorologie (ONM) of Algeria and Mauritania, respectively. This paper provides an overview of the observational program at the Fennec SS1, which provides both scientific insights into Saharan meteorology and a context for future planned research based on Fennec observations.
 The intense heating and the arid environment in the Sahara lead to one of the deepest boundary layers on Earth, with dry convection routinely reaching 5 or 6 km [Gamo, 1996; Cuesta et al., 2008; Cuesta et al., 2009]. Observations from AMMA in 2006 revealed , however, that summertime dry convection does not always reach these altitudes, even late in the afternoon, and the “Saharan Residual Layer” (SRL) can persist throughout the day over many hundreds of kilometers [Messager et al., 2010]. Mechanisms that support the persistence of the SRL have been hypothesized [Flamant et al., 2007; Marsham et al., 2008b] and investigated in models [Huang et al., 2010; Birch et al., 2012], but observations with which to test these hypotheses have been very limited.
 BBM is located close to the maximum mean aerosol optical depth (AOD) in summer [Prospero et al., 2002, Figure 1]. Wind-driven dust emission is a nonlinear process, commonly parameterized as a thresholded cubic function of friction velocity [Marticorena and Bergametti, 1995]. This makes small-scale high wind speed events potentially very significant, but the relative contributions of different meteorological mechanisms remain poorly quantified [Knippertz and Todd, 2012]. It is known that the diurnal mixing of low-level jets (LLJs) to the surface is the dominant mechanism in some regions, for example, in the Bodélé depression [Washington and Todd, 2005; Todd et al., 2008] and that this mechanism contributes farther west [Knippertz, 2008; BouKaram et al., 2009]. Monsoon surges have been observed to lead to dust uplift [BouKaram et al., 2008], and cold pool outflows from moist convection (“haboobs”) are generated when sufficient moisture from the desert margins allows storms to form [Flamant et al., 2007; Marsham et al., 2008a], although these are poorly represented in global models [Marsham et al., 2011a]. Boundary layer convection and dust devils are also known to generate dust uplift, although the overall contribution of these mechanisms is disputed [Cakmur et al., 2004; Koch and Renno, 2005; Marsham et al., 2008b].
 Dust in the Saharan boundary layer provides a radiative forcing that is known to be a source of uncertainty in models [Haywood et al., 2005; Rodwell and Jung, 2008; Lavaysse et al., 2011], and its influence extends far over the Atlantic as it is exported in the Saharan air layer (SAL) [Karyampudi and Carlson, 1978]. It is known that the interaction of dust with radiation means that including prognostic dust in numerical weather prediction (NWP) models improves forecasts in Europe [Tompkins et al., 2005], and many NWP centers are developing this aspect of their model. This makes observations from dust source regions even more important for model validation. Retrieval of dust properties from satellite observations remains problematic, especially over land, and validation of these algorithms is limited by a lack of observations from the central Sahara [Brindley et al., 2012]. Such observations are now required for assimilation into weather models that include dust as a prognostic variable.
 The observations made from BBM during the Fennec IOP allow insights into many of the processes and questions described above and raise further questions. The observations made are described in section 2. Data are presented in section 3, together with the meteorological context. Conclusions are summarized in section 4.
2 The Bordj Badji Mokhtar Supersite-1 During Fennec IOP, June 2011
 The Fennec supersite was deployed close to the existing ONM synoptic station at the BBM airport (Figure 1, 21.38°N, 0.92°E, ~420 m above mean sea level, with solar noon at 1157 UTC on 15 June 2011). Installation was begun slightly later than planned on 2 June 2011, and the site was run until the end of June. Instruments deployed (Table 1) included an instrumented 15 m mast (the “flux tower”), a Scintec MFAS phased array sodar, a HALO Photonics Streamline 1.55 µm Doppler lidar, Vaisala RS92 GPS radisosondes launched every 3–6 h, a Cimel Sun photometer, an aerosol sampling station (ASS), an inverse nephelometer, and a real-time absorption reflectometer. The flux tower was instrumented with 20 Hz sonic anemometers at 10 and 15 m. At 2 m, there were passively ventilated measurements of temperature and humidity, as well as pressure. A separate 2 m mast was used to measure upwelling and downwelling solar and longwave radiative fluxes (a Kipp and Zonen CNR4 radiometer) and ground heat flux at 10 cm. Internal temperatures from the radiometer are also shown, as these unventilated data are still useful for detecting temperature changes when the ventilated data are not available. The Nucleopore filters from the ASS will be submitted to gravimetric analysis, electron microscopy, elemental composition, and optical reflectance techniques. Nephelometer and absorption reflectometer data also require filter analysis, and this is still in ongoing.
Table 1. The Instruments Deployed at BBM During Fennec IOPa
All dates are in 2011. The flux tower was a custom-built unit, using Metek USA-1 sonic anemometers, an E-pluse ee04 humidity sensor, a Kipp and Zonen CNR4 net radiometer, and a Hukseflux ground heat flux sensor.
RS92 GPS radiosondes
3–6 hourly; surface to < 200 hPa
Also two radiosondes on 1 June.
University of Leeds, UK, with various components—see caption and text.
1 Hz, with 20 Hz sonic anemometers
6–30 June and ongoing
Some data gaps (see Figure 3). Ground flux calibration problem after 28 June.
Sodar (1650–2750 Hz)
300 frequencies per 10 min output period; 98 (10 m) range gates from 30 to 1010 m
2 June to 1 July
No sound shields before 6 June.
Doppler lidar (1.55 µm)
1.4 s averaged to 10 s; 316 (30 m) range gates from 105 to 9585 m; half-hourly wind profiles
2 June to 1 July
Data gaps from insufficient air conditioning before 13 June.
Calibration errors before 7 June.
Inverse nephelometer (670 nm)
Laboratory for Aerosols, Clouds and Optics (LACO) at the University of Maryland, Baltimore County (UMBC)
4 s averaged to 1 min and smoothed to 30 min
Analysis is ongoing.
Real-time absorption reflectometer (red, green, and blue)
Some problems before 22 June. Analysis is ongoing.
Automated aerosol sampling (ASS) station
Two size ranges of <1.5 and 1.5–10 µm on Nucleopore filters
 Table 1 shows the periods when data were available. Many instruments experienced problems with overheating, showing the challenges of working in this harsh environment; these problems were largely solved by 17 June. In particular, additional air conditioning was required to stop the lidar overheating, and the voltage regulators of the flux tower often overheated at around 11 UTC before 17 June. The data suggest that some flux tower dropouts may be due to the strong winds and turbulence from the most powerful haboobs (e.g., 17 June). Sodar sound screens were not deployed until 6 June, but data from before this time were found to be of good quality. Due to problems with the heater on the ground heat flux plate, ground heat flux data are not shown after 29 June. There were problems with an air leak on the reflectometer before 22 June, and the data are still being analyzed, as a result, reflectometer data are not shown.
2.1 Eddy Covariance and Dust Uplift
 Data (20 Hz) from the sonic anemometers (measuring three-dimensional winds and virtual temperatures) were used to calculate 15 min values of sensible heat and momentum fluxes by eddy covariance. This 15 min period was a compromise between desired high temporal resolution and increasing errors for shorter periods [Lenschow et al., 1994]. The eddy correlation analysis therefore gives 15 min values of friction velocity, but instantaneous values of surface stress cannot be calculated. The 15 min averaging period also gives unrealistic values of friction velocity at sharp fronts, where there are large changes in mean wind through the period (see discussion of Figure 5d in section 3.5).
 From 22 to 28 June (continuous data with an integer number of diurnal cycles), the mean net residual between the net observed radiative fluxes, ground heat flux, and sensible heat flux calculated by eddy correlation was −0.63 W m−2, suggesting a very small net latent heating of the atmosphere over this period (as expected) and good accuracy in the time-averaged eddy covariance data sensible heat fluxes.
 For a fixed location such as BBM over the duration of June 2011 (when few changes in land surface characteristics are expected), dust uplift is largely a function of the surface stress, which is often expressed using friction velocity. Uplift is normally parameterized as a thresholded cubic function of u*, e.g., proportional to [Marticorena and Bergametti, 1995]
Marsham et al. [2011a] referred to this wind-dependent component of the parameterized dust uplift as the “uplift potential” and calculated it using 10 m wind speed (u) instead of u* (since u* was unavailable to them). Using 15 min u* to calculate uplift potential misses uplift potential from events that give large surface stresses for a short period: such events can give a 15 min mean u* that is less than u*t even though u*t is exceeded for a short period. Furthermore, due to the nonlinear nature of uplift potential using 15 min u*s also underrepresents events with highly variable surface stresses through any 15 min period. These problems are avoided by simply using u, which is available at higher frequency. Section 3.5 further discusses the relative merits of using u* and u for calculating uplift potentials for the BBM data.
2.1.1 Other Data Sets
 In this paper, we also use false color SEVIRI (Spinning Enhanced Visible Infrared Imager, carried by Meteosat) imagery [Lensky and Rosenfeld, 2008]. In addition, Figures 5 and 8 show the fraction of the nine SEVIRI pixels closest to BBM masked as cloudy by the European Organisation for the Exploitation of Meteorological Satellites' MPEF (Meteorological Product Extraction Facility) cloud mask algorithm (referred to as the “SEVIRI cloud fraction”). By visual inspection of imagery, it was determined that these sometimes incorrectly mask thick dust as cloud. Finally, we use operational analyses from the European Centre for Medium-Range Weather Forecasting (ECMWF) at 1° resolution to provide a synoptic scale context of the data from BBM.
3 Analysis of Observations From Fennec Supersite-1 at BBM Made During the Fennec IOP
3.1 Synoptic Scale Evolution
 Figure 1 shows the mean synoptic scale condition during June 2011 at 925 hPa from ECMWF analyses. BBM is on the western edge of the mean location of the SHL. A dew point of 14°C is often used to define the leading edge of the monsoon, i.e., the intertropical discontinuity (ITD) [Eldridge, 1957; Buckle, 1996]. Figure 1 shows that the ITD was, on average, approximately 4° to the south of BBM but did reach BBM during June 2011. BBM is very close to the maximum in dustiness as defined by the aerosol index from the ozone monitoring instrument (OMI), suggesting that dust uplift processes observed at BBM are significant for dust export from North Africa.
 May to July rainfall anomalies [from the National Centre for Environmental Prediction (NCEP) Climate Prediction Centre, produced by merging gauge and satellite data] indicate that the June 2011 West African Monsoon rains in the Sahel were not exceptional (not shown). Figure 2 shows that June 2011 at BBM can be spilt into three main periods. Before 8 June, the center of the SHL was close to BBM, with BBM being affected by moist south-westerly monsoon winds, particularly toward the end of this period (Lavaysse et al.  used the thickness to define the SHL and the West African HL, but here we simply define the SHL position as the location of minimum near-surface pressure in the analysis). The second period, from 8 to 12 June, was characterized by the SHL being far to the east, with BBM in dry northerly Harmattan winds and Fennec supersite-2 strongly influenced by maritime inflow (Todd et al., submitted manuscript) . An upper level trough was located to the northwest of BBM at this time. The third period, from 13 June onward, was characterized by moist monsoon winds and dry northerlies alternating on periods of approximately 3 days; these are consistent with African Easterly Wave activity, which is known to favor dust emission in this region [Knippertz and Todd, 2010]. The SHL was observed to repeatedly propagate westward during this period, with pressure minima at BBM on 20 and 26 June. From 25 June, an upper level ridge built northward over BBM, moving westward with the SHL. Figure 3c shows that dust was largely observed in the two moist periods, rather than the dry period from 8 to 12 June. There are few data from the Fennec supersite from the first of these two periods (Table 1), so the analysis here is largely focused on the last two periods: northerly winds from 8 to 12 June and alternating flow regimes thereafter.
 Radiosondes and the flux tower at BBM show a dry lower boundary layer (~900 hPa) from 8 to 12 June, with boundary layer water vapor mixing ratios (WVMRs) not exceeding 5 g/kg (Figures 3, 4, and 5). The moist period from 13 June is characterized by five sudden moistening events seen in the radiosondes and flux tower data (around 00 UTC on 13, 17, 21, 25, and 29 June, shown by arrows in Figure 3). Both of the moist periods (before 8 June and after 12 June) are associated with southerly winds (Figures 2, 4, and 5).
3.2 The Boundary Layer
 The radiosonde profiles (Figure 3a) show the characteristic deep, almost dry adiabatic region from the surface to around 600 hPa. The boundary layer depth was diagnosed as where the virtual potential air temperature first reaches 0.5 K above the value at 930 hPa (Figure 3a, lower solid line) and the WVMR first falls to 0.5 g kg−1 below the value at 930 hPa (dashed line). The lowest level of 90% relative humidity was used to indicate likely cloud base (upper solid line). Despite the limitations of these simple algorithms, the two boundary layer heights normally agree to within 20 hPa. They are therefore useful diagnostics indicative of the mixing, but care should be taken in their use, especially when complex layerings are present. The deep daytime boundary layer is seen to regularly reach or almost reach cloud base, consistent with the convective cloudiness often observed in SEVIRI.
 After 13 June, at some time in the afternoon, the atmosphere is generally almost well mixed, with no residual layer, from the surface to 600–450 hPa. However, before 13 June, the diagnosed boundary layer depth only reaches around 750 hPa. Before 13 June, a moist residual layer of generally southwesterly winds is seen to overlay the drier boundary layer, which has northerly winds (Figures 3b and 4), similar to the aircraft transect shown by Messager et al. . Figure 6 shows two afternoon tephigrams from before and after 13 June. The latter shows a deep, well-mixed boundary, while the earlier shows the moist SRL. However, even the deep boundary layer from 15 June (Figure 6b) is not absolutely well mixed; above 700 hPa, the wind changes and the WVMR become increasingly variable, indicative of moisture variations from the large plumes and dry entrained air expected at these altitudes close to the boundary layer top.
 Intriguingly, there is a diurnal cycle in the WVMR in the SRL before 13 June in Figure 3b, which may be real or may be a function of solar heating of the radiosondes. Between 8 and 13 June, the lidar (Figure 3c) shows dust in the moist residual layer that persists over the convective boundary layer (CBL). The weak lid of around 2 K seen in Figure 6a is clearly sufficient to inhibit mixing of dust and water between these layers.
 Figure 7 shows back trajectories from BBM produced using ECMWF analyses. Due to the strong boundary layer convection and mixing over the Sahara, caution must be applied when interpreting these, but they show a strong contrast between 11 and 15 June, i.e., from when a persistent SRL was present and when the boundary layer was well mixed. On 11 June, Figure 7a shows that the air at low levels originated from the cold Atlantic, with the 4 day trajectories descending to 950 or 850 hPa from around 850 or 580 hPa, respectively (air above this had been over the hot continent for all 4 days). This is consistent with the strong maritime influence observed at Fennec supersite-2 during this time (Todd et al., submitted manuscript). On 15 June, Figure 7b shows that the low-level air is more continental, with some Mediterranean influence, with trajectories again descending to 950 or 850 hPa from around 650 and 620 hPa. It appears that, on 11 June, the Atlantic air is sufficiently cold for the boundary layer convection to have not yet mixed it with the SRL above. Analyses show that, for 8–11 June, the SHL was to the east over Niger, with a trough close to the Atlas Mountains allowing this ventilation, with the trough linked to a midlatitude disturbance visible in 300 hPa winds (Figures 2a and 7a). From 12 to 15 June, the SHL was still to the east, but this trough was not present, removing the ventilation by Atlantic air (Figures 2 and 7b). Overall, the presence of a midlatitude trough to the northwest of BBM from 4 to 9 June appears to have ventilated the region to the west of BBM, resulting in the SHL being displaced eastward on 6 June and only moving westward again once the influence of the trough had been removed.
 It is not yet clear how well global models can capture the weak lids and layerings observed at BBM and their impacts on dust transport, although Met Office operational forecasts did show a 1–2 km deep CBL extending westward from BBM on 11 June that deepened to around 5 km by 15 June (not shown), consistent with the model capturing the Atlantic ventilation shown in Figure 7. Future Fennec research will address these boundary layer processes in detail, making use of the aircraft observations.
3.2.1 The Diurnal Cycle in the Boundary Layer
 The strong diurnal cycle in heating and cooling (section 3.6.1) gives a mean diurnal cycle in 2 m temperature of 30°C–42°C, with a maximum around 15 UTC and a minimum close to 06 UTC, just after sunrise at around 0515 UTC (Figure 8b). The expected atmospheric tide is seen in the pressure data [Pugh, 1987]. As a result of the dry atmosphere and intense heating, a deep boundary layer is observed, and at 15–17 UTC, potential temperature is almost well mixed to 550 hPa (Figure 9a). The boundary layer depth diagnostics used in Figure 3 show agreement with this value, with a mean well-mixed depth of 5000 m at 15–18 UTC (550 hPa). The lidar and sodar (Figures 10 and 11) show the magnitude of daytime vertical winds increasing upward, as expected, although both suffer from sampling biases at upper levels (which are discussed at the end of this section). Maximum values of the magnitude of the vertical winds in the boundary layer occur at 13–15 UTC, when surface temperatures (but not surface fluxes) are largest. There is evidence of elevated maxima in turbulence at 17–22 UTC in Figure 10a, which may be downdraughts generated by evaporating precipitation.
 WVMR is almost well mixed from 15 to 18 UTC, except for a drying toward the upper 1000 m of the boundary layer (Figure 9b). Surprisingly, at 15–17 UTC, WVMRs are lower in the lowest 1000 m than in the 1000 m above: this may be an instrument error, a differential advection, or from the evaporation of precipitation that does not reach the ground (it is still present if only data from after 13 June are used, so it is not just a result of the residual layer present before that date). The diagnosed boundary layer depth decays from 18 to 21 UTC, but the depth of WVMRs exceeding 6 g/kg increases to 21 UTC and relative humidity is maximized at 21–00 UTC (Figure 9b, solid lines), consistent with the cloud maximum at 18–21 UTC (Figure 8a).
 Overnight, cooling is observed below 2500 m (Figure 9a). Wind speeds and WVMRs are increased in this deepening cold layer, with a pronounced wind speed maximum below 1000 m (Figures 9, 10, and 11) from the nocturnal LLJ (discussed in detail in section 3.3.1). Radiosondes show a moistening from 6 to 8 g kg−1 from 15 to 09 UTC, which corresponds to a wind speed increase from 4 to 13 m s−1 from 15 to 06 UTC (Figure 9). Flux tower data show WVMRs increasing from 5.5 to 6.5 g kg−1 from 15 to 08 UTC. This moistening is due to nighttime advection of moist air facilitated by the sudden removal of turbulent drag at sunset and the LLJ [Parker et al., 2005] and moist haboobs (section 3.4, convective outflow start times are shown by vertical blue lines in Figure 8e). After sunrise the boundary layer deepens, mixing the moisture upward, eroding the low-level stability, and mixing momentum from the LLJ to the surface (section 3.3.1).
 It should be noted that, although time series of lidar and sodar winds agreed with those from radiosondes, their sampling characteristics (Table 1) affected their perceived diurnal cycles (Figures 9, 10, and 11). Only radiosondes sample all levels equally often; the sampling of any level by the sodar or lidar depends on sufficient backscatter being detectable, affecting the mean diurnal cycle at upper levels. The lidar gave consistent wind data for levels below 500–1000 m (Figures 3c) and the sodar below around 300 m [Figure 11b shows that the sodar's reliable layer of high backscatter (greens) varies from around 150 m at night to 300 m at 09 UTC]. Therefore, each instrument provides additional insights into the winds and their diurnal cycle, but their sampling must be accounted for. In particular, the range of high sodar backscatters does not capture the full vertical extent of the nocturnal LLJ, which the lidar and radiosondes both show has peak winds at around 300 m (Figures 9 and 10), but the sodar does provide good vertical resolution in the winds close to the surface, from a level of 30 m upward.
3.3 Boundary Layer Jets
3.3.1 The Nocturnal Low-Level Jet
 Figures 9, 10, and 11 from radiosondes, lidar, and sodar show the pronounced wind speed maximum below 1000 m from the nocturnal LLJ, which brings moist air to BBM. The lidar (Figure 10) reveals turbulence, presumably generated by wind shear, at the top of the LLJ layer (around 1100 m) from 05 UTC until around 10 UTC, when developing boundary layer convection reaches this level (Figure 10a). This shear-induced turbulence will restrict the development of the LLJ. Figure 12a (dashed line) shows the classic near-circular clockwise diurnal rotation of winds in the LLJ [Blackadar, 1957; van de Wiel et al., 2010]. Figure 12b shows the anticlockwise rotation at low levels, which is consistent with the theory proposed by van de Wiel et al. , which was not captured well by the radiosondes, which have limited resolution at these low levels. Winds from 06 to 15 UTC—approximately the period when dry convection is most active at 100 m—are relatively similar to each other, and between 15 and 06 UTC, the winds describe a near-circular oscillation.
 From 08 to 11 UTC at 100 m, the sodar shows an area of high winds extending toward the surface (Figure 11a), which corresponds with a period of increased sodar backscatter from the moist dusty air. The mixing of the LLJ momentum to the surface gives a strong peak in 10 m winds at 8–10 UTC (Figure 8) [May, 1995]. There are then separate peaks in wind and uplift potential at 1115 UTC, which tail off toward 16 UTC. The wind peak at 1115 UTC disappears if data from 29 June is excluded, since this day had exceptionally high winds for this time of day (11–16 UTC), with no flux tower data before 11 UTC.
 To examine the LLJ breakdown, it is useful to look at the 10–15 m shear seen in the flux tower data (Figure 8, blue line). This shear is strong in the nocturnal boundary layer, as expected, and weaker during the day due to dry convection. The 15 m wind speeds are actually 0.1 m s−1lower than 10 m wind speeds from 0920 to 1040 UTC, but this is a likely random error (an uncertainty of just 0.07 m s−1 in each wind speed is sufficient for this). Regardless of this, 1040 UTC is approximately the time of the reversal of the sign in the change of the 10–15 m wind shear with time: between 0630 and 1040 UTC, the jet is breaking down and shear decreases; after 1040 UTC, the shear is increasing. Most momentum from the LLJ appears to have been dissipated by 11 UTC. However, Figure 9c shows that the jet persists until 12 UTC, but not 15 UTC: Figure 11a shows the LLJ largely dissipated by 12 UTC. Therefore, in Table 2 and Figures 3–5, the LLJ period is denoted as 06–12 UTC, with approximately 30% of uplift potential occurring during this time (Table 2). At 0–200 m, the sodar shows maxima in the vertical winds at 06 and 11 UTC, before and after the main breakdown of the LLJ (Figure 11c), and the lidar shows vertical winds increase suddenly at 11 UTC after the LLJ breakdown (Figure 10a).
Table 2. Fractions of Uplift Potential and Nephelometer Scattering Occurring During Different Periods, Defined by Either Time of Day (“Nocturnal,” “LLJ,” and “PM”) or Mechanism (“Convective Outflows”)a
Fraction of Dust Uplift Potential (%)
Fraction of Scattering (%) [When Wind Data Exist]
Uplift potential is calculated from 1 s (10 m) wind, using four threshold velocities. Fraction of scattering is calculated both for all nephelometer data and for only periods when wind data exist and so uplift potentials are defined (in ). For uplift potentials, 0.3%–0.5% is in both “Convective outflows” and “Early PM.” 0.0% is in both “Convective outflows” and “LLJ.” “Convective outflow” periods are defined in text (Figure 3, blue bars). For “Early PM” (i.e., 12–16 UTC) a significant fraction occurred on 1 day (29 June) so statistics with this day excluded are shown in round brackets.
“Nocturnal”: 18–06 UTC
“Low-level Jet (LLJ)”: 6–12 UTC
“PM”: 12–18 UTC
“Early PM”: 12–16 UTC (excluding the 29th)
8.3 (4.5 )
 Figure 4a shows that the LLJ wind speeds around 930 hPa are often comparable with or stronger than those in the free troposphere at 300 hPa. The typical time of momentum from LLJs reaching the surface (06–12 UTC) is indicated by red bars above the x axes in Figures 3, 4, and 5. In the radiosonde (Figure 4), lidar, and sodar winds (not shown), the LLJ is strongest on 16 June, when dry northeasterly winds reached 25–30 m s−1, but unfortunately little flux tower data were obtained (what flux tower data are available from 16 June shows an 18 m s−1 wind speed at 10 m). Other strong jets occurred on 25 June (a moister southerly, 12 m s−1 at 10 m), 26 June (a moist southerly, 15 m s−1), and 29 June (a moist southeasterly, 16 m s−1 at 11 UTC). These strong jets are associated with strong pressure gradient close to the ITD. The three moist jets are all associated with overnight monsoon surges moistening BBM (Figures 2 and 3b), with convective activity in the preceding night (section 3.4). After each jet, a corresponding increase in AOD was observed at BBM of approximately 2–3, 3.5–5.5, 3–4, and 3–6 on 16, 25, 26, and 29 June, respectively. In all these events, the momentum reaches the surface at approximately 08–12 UTC, rather than at night (as observed in BouKaram et al. ). It may be that a stronger nocturnal inversion at BBM compared with the moister Sahelian region discussed in BouKaram et al.  inhibits strong nocturnal winds from monsoon surges at BBM.
 Data from 25 and 26 June are shown in Figure 13. Moderate (8 m s−1) dusty monsoon winds precede a 06 UTC wind speed minimum on 25 June, before the LLJ reaches the surface, with a maximum of 10 m wind speed at 0930 UTC (~12 m s−1). Both the lidar and nephelometer data show much dust is transported within the moist air in this case (with weak fronts at 02 and 06 UTC), rather than being locally generated. 26 June has weaker monsoon winds and a stronger jet (~ 15 m s−1), perhaps because less momentum has reached the surface during the night in this event; this event generates significant uplift, with nephelometer data closely following uplift potentials. The winds are very gusty in the afternoons of both days. 29 June also deserves mention: convective clouds were over BBM for the whole day, but no cold pool was visible in SEVIRI until 19 UTC. Flux tower data (available from 11 UTC) or lidar data also did not show cold pools until 19 UTC. As already discussed, this day was associated with a pronounced monsoon surge. It had unusually strong afternoon winds (10 m s−1 at 15 UTC), which contributed a significant fraction of the total uplift potential observed at this time of day (Table 2). Figure 13f shows narrow dusty columns above the surface at around 14 UTC; these are presumably dusty convective plumes or dust devils (similar to Ansmann et al. ), but the small fraction of uplift potential occurring between 12 and 16 UTC (Table 2) suggest these features are not a dominant source of dust uplift at SS1 in the Fennec IOP period.
3.3.2 The Saharan Upper Boundary Layer Jet
 As well as the nocturnal LLJ, the data reveal a second jet between approximately 3000 and 5500 m, which we refer to as “the Saharan upper boundary layer jet.” Between these levels, from 00 to 09 UTC, radiosondes show a gradual decrease in WVMR and relative humidity (Figure 9). This drying is consistent with the minimum in cloudiness at 06 UTC (Figure 8a). Wind speeds in this layer are at a minimum from 10 to 18 UTC, presumably since they are inhibited by boundary layer convection at these times. After 18 UTC, these winds accelerate, bringing in dry air until 09 or 10 UTC, when these levels start to be remoistened by boundary layer convection. Figure 12a (solid line) shows a clockwise near-circular diurnal rotation at 650 hPa (approximately 4000 m) around the mean east-northeasterly wind created by the flow around the high pressure found at this level (the northern limit of the African Easterly Jet, not shown). This is consistent with an inertial oscillation, similar to that driving the LLJ around the SHL (the reason for the outlier at 06 UTC in Figure 12a is unclear). In the Sahel, the African Easterly Jet is affected by boundary layer convection [Kalapureddy et al., 2010]. The convection in the Sahara is even stronger, and the depth of the Saharan boundary layer means that winds at around 650 hPa are still affected by convection, despite their altitude, so feel drag from the surface, allowing a Saharan upper boundary layer jet to form.
3.4 Cold Pool Outflows (Haboobs)
 Cold pool outflows from local or remote moist convection were regularly observed. These varied from events that lasted less than 10 min, which were observed when SEVIRI showed convective clouds over BBM, to outflows that were remote from the source and spread over hundreds of kilometers and could be clearly seen in SEVIRI imagery due to lifted dust and sometimes arcus clouds.
 Figures 14a, 14c, and 14e show an example of a short-duration outflow event. Clouds are seen over BBM in the imagery (Figure 14a). At 02 UTC, there is a sudden increase in pressure (~0.5 hPa), temperature (~2 K), and WVMR (~0.5 g kg−1). The nephelometer data are unclear, but at this time, there is a sudden appearance of dust in the lidar data, which show disturbance of the stratification, with time this dust appears to be lofted above a cleaner layer, presumably the nocturnal boundary layer: the short duration of the appearance of dust at low levels in the lidar data consistent with the short duration of the high winds and uplift potential. This appears to be a warm microburst type of event. The deep dry boundary layer of the Sahara means much, and often all precipitation evaporates before it hits the ground. The resultant cold moist air has little to inhibit its descent, and if it overshoots on its descent, it is seen as a warm moist wind gust. A second, much weaker event is seen at 03 UTC on 7 June (Figure 14c).
 A much larger event was observed on 29 June (Figures 14b, 14d, and 14f). Although convection was again close to BBM in this case, the haboob could be seen in the satellite imagery (Figure 14b). This imagery suggests that the first moistening at ~1930 UTC was an outflow, but this is not seen clearly in wind speed or the lidar data. The main haboob arrived at ~2130 UTC. This gave approximately a 4 K temperature decrease, a 2 hPa pressure increase, and a 6 g kg−1 WVMR increase. Wind speeds increased from 5 to 23 m s−1. There was a slight double front in all these variables (and in the lidar data), suggesting a complex structure to the head of the cold pool, perhaps a bore preceding the cold pool, but this might be expected to not give such a significant cooling [Marsham et al., 2011b]. The lidar and nephelometer show very high dust loadings within the cold pool, much larger than in the two LLJ events shown in Figure 13 (with loadings decreasing after approximately 6 h). The dust rapidly attenuates the lidar, making determination of the depth of the cold pool from the lidar data problematic. Again nephelometer scattering values closely follow the dust uplift potentials.
 The characteristics of most cold pool events lie between the two events shown in Figure 14, with the strongest wind speed events at BBM observed from cold pools at ~1715 UTC on 17 June (~25 m s−1) and 1830 UTC on 22 June (25 m s−1). Microburst-type events tended to occur around the evening, when clouds over BBM were most developed (section 3.6.1 and Figure 8a), while larger haboobs from the Sahel tended to arrive in the evening and in the night. The determination of “convective outflow periods” when BBM was affected by active downdraughts or cold pool outflows (blue bars below x axes on Figures 3, 4, and 5) was done by inspecting all available data and imagery and is subjective. Some events were very clear, i.e., SEVIRI showed the haboob and it was detected in surface and lidar data. Others were less clear, and some have probably been missed, especially events under cloud when flux tower and lidar data were not available (note, however, that Table 2 only includes periods when flux tower data were available by definition). Figures 13b and 13d show an example of a further, slightly uncertain event at around 0145 UTC on 26 June; convection is immediately southeast of BBM, and there is a very rapid 0.5 hPa jump in pressure and a sharp wind speed spike, with sudden wind direction changes, similar to, but less distinct than, those in Figure 14c, with lidar showing high backscatter. Figures 13a and 13c show that at 0230 UTC on 25 June convection was immediately northwest of BBM, but the ~10 m/s winds are not associated with sharp pressure, temperature, humidity, or wind direction changes, and therefore, the event appears to be a monsoon surge, which had uplifted dust (see lidar, Figure 13f), rather than a cold pool outflow. In general, it is much easier to define the start of convective outflow events than their end, but a less conservative definition of cold pool periods than that used here only increased the cold pool fractions in Table 2 by a few percentage, since the strong events that give large uplift potentials are also those that are most clearly defined.
 The five sudden moistenings observed at BBM (marked on Figure 3) each correspond to the arrival of outflows from convection (blue bars below the x axes of Figures 3, 4, and 5). Such cold pools are often observed to form part of such surges [e.g., Flamant et al., 2007; BouKaram et al., 2008; Marsham et al., 2008a]. The most dramatic moistening at BBM occurred at around 06 UTC on 13 June and resulted from a large haboob arriving from an MCS that initiated to the southeast over the Aïr Mountains and had WVMRs increasing from ~2 to ~8 g kg−1, AOD ~1 to ~4.5 (Figure 5). This was not the first monsoon influence at BBM that summer; there were haboobs in the first moist period in June (before 8 June) and there was a large haboob on 29 May, but haboobs were much more frequent after 13 June than before. Overall, WVMRs from radiosondes integrated from the surface to 750 hPa show that 34% of the total moistening occurred during periods of cold pool outflows and 12% of the drying (drying can occur as though cold pools are normally moist, parts are drier than others, e.g., Figure 14d, and drying can occur above the cold pool). This does not actually show that cold pools were responsible for a significant fraction of the moistening, which may be largely a result of the monsoon flow in which they are embedded, but neither does this allow that hypothesis to be rejected.
3.5 Dust and Dust Uplift
 The nephelometer gives a measurement of scattering at 2 m, the lidar backscatter at 105 m to around 3000 m, and the Sun photometer gives the AOD (Table 1). All three instruments show the most dust in the monsoon-affected periods (Figures 3c and 5e), in particular after 13 June (there are variations in nephelometer scattering between 6 and 13 June, but these are too small to see in Figure 5e).
 The nephelometer, lidar, and Sun photometer give measures of dustiness, as distinct from dust uplift, with dustiness affected not only by any ongoing local dust uplift but also by previous local uplift and dust advected over the site. We therefore use uplift potential [Marsham et al., 2011a] as a measure of the relative contribution of various meteorological mechanisms and events to dust uplift. As described in section 2.1, uplift potential can be calculated using 10 m winds (u) or friction velocities (u*). For winds greater than a typical threshold (6–9 m s−1) [Chomette et al., 1999], flux data showed that u* is almost proportional to u, so u gives a useful estimate of u*, but at a higher time frequency than eddy covariance estimates of u* (not shown). We therefore use 10 m winds to calculate uplift potential in the remainder of this paper, since (i) Figure 5d shows the unrealistically large values of friction velocities derived from eddy covariance at fronts, where there are large changes in mean wind, and (ii) there are no high-frequency measurements of friction velocity, which biases uplift potentials (section 2).
 The use of uplift potential approach is supported by data from the lidar and nephelometer, with Figure 5e showing that high values of uplift potential lead to high scattering from the nephelometer and Figures 13 and 14 showing that, when dust is locally generated, lidar and nephelometer data are consistent with uplift potentials calculated with ut = 7 m s−1 (Figures 13d, 13f, 14d, and 14f). Figure 15 shows the probability density functions of 10 m winds and lidar backscatter at the lowest range gate. Low winds are associated with a range of backscatter from dust uplifted previously or remotely, and high winds are associated with high backscatter from active or remote dust uplift: there is a decrease in the variance in the backscatter as wind speeds increase and the backscatter is increasingly controlled by freshly uplifted dust. Above approximately 8 m s−1, there is a distinct projecting “arm” of high winds and high dust, which must be from fresh dust uplift. The dashed line in Figure 15 shows the mode of the backscatter, for 1 m s−1 wind speed bins, showing a sharp increase in backscatter for winds above 8 m s−1, although there is evidence of increased backscatter above 5 m s−1 (the line dips down at 9 m s−1 rather than going up to the secondary peak, which is continuous with the projecting “arm” of high winds and high backscatter). We expect dust emission to scale with uplift potential, which is shown by the five red lines, each of which uses a different threshold velocity and the same arbitrary scaling. All the red lines in Figure 15 lie within the “arm” of high winds and high backscatter, with thresholds of 6 or 7 m s−1, perhaps giving the best fit. However, due to the uncertainties in ut, Table 2 uses a range of threshold velocities.
 Figure 5e shows that uplift potential is more sporadic than wind speed, as expected from the nonlinear nature of dust uplift, and consistent with the large time variations in scattering observed by the nephelometer. Uplift potentials are much greater in the moist period after 13 June, which was much dustier. Many uplift events occurred between 0630 and 12 UTC (red bars below plot). This timing is consistent with an LLJ mechanism and the strong LLJ events on 16, 25, 26, and 29 June are clearly seen in both uplift potentials and nephelometer data. Many of the largest uplift potentials and nephelometer scattering values are associated with cold pool outflows (blue bars). The large cold pool events on 13, 17, and 19 June give large increases in AODs (approximately 2–7, 2–7, and 4–6.5), but AOD changes are hard to assess, as often BBM is either cloudy or dark for these events. Overall, Table 2 shows that around 50% of 10 m wind uplift potential is from convective outflow periods and 33% of nephelometer scattering; the smaller percentage for the scattering compared with the uplift potential is consistent with the dust uplifted in the active cold pool periods continuing to be observed by the nephelometer after the period of active uplift within the cold pools (see also section 3.5.1). Around 30% of scattering and uplift potential occurs between 06 and 12 UTC, which is consistent with LLJs (Table 2 and section 3.3.1). There were limited 10 m wind data from the strong LLJ event on 16 June (Figure 5) for inclusion in the statistics in Table 2; however, Figure 5 shows this event led to only a modest AOD increase and this event was observed by the nephelometer.
 The fact that haboobs occur in monsoon surges means that some high winds are expected even without downdrafts from convection [e.g., BouKaram et al., 2008] (Figure 14c) and it is not possible to fully separate monsoon winds and convective outflows in this paper, and indeed this may require numerical modeling [e.g., Marsham et al., 2011a]. However, cold pools provide the largest AODs, and the two largest uplift potential events were haboobs, with 17% and 10% of the total uplift potential each (with ut = 7 m/s). In other words, single rare haboob events contribute substantial fractions to the total dust uplift, contributing to large variability in dust on all time scales.
3.5.1 The Diurnal Cycle in Dust and Dust Uplift
 Fifty-one percent to 55% of uplift potential and 50%–52% of scattering observed with the nephelometer occur at night (Table 2). Comparing the blue and black lines in Figure 8e shows that this is almost all from haboobs, although monsoon surges contribute a small fraction (e.g., Figure 13). There is a minimum in uplift potential at 06 UTC, after most haboobs and when the nocturnal inversion is strongest, inhibiting the mixing of momentum from LLJs and haboobs to the surface. The daytime mixing of the LLJ momentum to the surface gives a strong peak in uplift potential at 08–10 UTC (Figure 8), which as discussed above is actually continuous with the peak at 1115 UTC, giving 30% of uplift potential between 06 and 12 UTC. Between 1115 and 16 UTC, uplift potential declines and, at this time of day, contributes little to the total, especially if the high winds on 29 June are excluded (Table 2 and discussion of this day in section 3.3.1). The gradual decline in uplift potential through the day is consistent with boundary layer convection inhibiting synoptic scale winds [Parker et al., 2005] and shows that dust devils and convective plumes do not dominate dust emission in this region.
 After 16 UTC, there are various high uplift potential events with high variance (Figure 8e). This is the period when cold pools were noted (cold pool start times are shown by blue vertical lines, all between 1430 UTC and 06 UTC). The sporadic nature of the cold pools gives a large variance in uplift potential through this period. The time of day is a useful separator for LLJ and cold pools in BBM data, with no cold pools arriving at the LLJ time (although some persisted through it; Table 2). Individual peaks in uplift potential line up with haboob start times (clearest at 19–20 UTC and 22–23 UTC) corresponding to peaks in pressure and WVMR. This supports the hypothesis [Flamant et al., 2007; Cuesta et al., 2010; Marsham et al., 2011a; Marsham et al., 2013] that cold pools play a significant role in transporting moisture and cold air into the SHL (though much of the cold air transport is masked at the surface by the stable nocturnal boundary layer). Most cold pools start between 17 and 00 UTC (when observed uplift potentials are larger), rather than 00–06 UTC. In this region, models with parameterized convection have large uplift potentials from 00 to 06 UTC, while models with explicit convection have large uplift potentials from 15 to 00 UTC [Marsham et al., 2011a]. This supports the hypothesis that the cold pools themselves generate much of the uplift potential, rather than merely the monsoon flow in which they are embedded (section 3.4).
 The diurnal cycle in scattering measured by the nephelometer is consistent with the diurnal cycle in uplift potential, given that the nephelometer is affected by previous, possibly remote dust uplift, as well as ongoing local dust uplift. Both uplift potential and scattering have peaks at around 08 UTC from the LLJ and at night from haboobs. However, the LLJ scattering peak precedes the peak in uplift potential by approximately 45 min, with scattering decreasing after 0730 UTC. This difference is not a result of the slightly different periods sampled by the instruments (not shown) and is probably because the boundary layer deepens after this time, entraining cleaner air from above. At night, the scattering lags the uplift potentials, consistent with scattering from the accumulated preceding uplift. In particular, the weaker cold pools starting after 04 UTC have high scattering, considering their low uplift potentials, which is consistent with the observations of dust that was uplifted into these cold pools earlier, when they were more active. Overall, this gives a smaller fraction of scattering compared with uplift potentials within active cold pool periods, but similar fractions occurring between 18 and 06 UTC, when cold pools dominate uplift, and similar fractions between 06 and 12 UTC, when LLJs dominate (Table 2). This supports the use of uplift potentials for estimating the contributions of these two mechanisms to dust uplift. The diurnal cycle in the AODs shows a maximum at 13 UTC (Figure 8e). AODs are a complex function of local dust uplift and dust transport at all levels, and it is very difficult to disentangle these contributions from data alone. Residual cloud-masking errors may also contribute to AODs in the afternoon and in the evening.
3.6 The Surface Energy Balance
 Figure 5a shows the time series of shortwave, longwave, sensible, and ground heat flux at 10 cm. The maximum solar downwelling radiation observed, of 1351 W m−2 over 1 min, was on the low AOD 7 June, which had variable cloud. The minimum daily maximum recorded was around 665 W/m2 over 1 min on the dusty 17 June, which was cloud-free at midday (SEVIRI masks dust on this day as cloud), but no radiation data were available after 11 UTC on 13 June, which was a cloudier day and just as dusty. Figure 5a shows that net radiation (purple line), which is largely converted to sensible heating of the atmosphere (orange line), depended largely on shortwave down (red line). After 17 June, smaller values of net radiation were observed, due to reduced shortwave down due to dust (AODs in Figure 5e) and cloud (SEVIRI cloud mask in Figure 5a). Before 13 June, variations in shortwave down and hence net radiation were dominated by cloud due to sustained low AODs.The magnitude of these variations are hard to assess, however, since only 9 and 10 June have complete radiation data with AOD observations.
 Figures 16a–16c show the daily mean observed downwelling shortwave irradiation at the surface for days with Cimel data and complete daytime radiation data (9, 10, 17–27, and 30 June). The level 1.5 Cimel data give AODs when its cloud mask determines the sky to be cloud-free [Smirnov et al., 2000], so the number of Cimel observations in a day gives some measure of cloud fraction through the day. The daily variations in downward shortwave can largely be explained by the daily mean AOD (Figure 16a, correlation of −0.96). There is also a strong anticorrelation (−0.88, Figure 16b in black) between downward shortwave and the SEVIRI cloud mask fraction (this was similar if the SEVIRI cloud fraction was weighted by the idealized unpertubed shortwave at that time of day) and a strong correlation with the number of Cimel observations (0.79, Figure 16a in gray). The correlation between AOD and the SEVIRI cloud mask is 0.82 (not shown). Figure 16c (in black) shows that the correlation between AOD and number of Cimel observations is −0.64. Therefore, Figures 16a and 16b (black) are largely showing the same correlation: dusty days also tend to be cloudy, and the SEVIRI mask can mask thick dust as cloud. To try to circumvent this issue, the difference between a linear fit between daily mean downward shortwave and AOD and the actual downward shortwave is plotted in Figures 16b and 16c (both in gray). We expect this difference to largely depend on cloud. This difference is plotted against the mean cloud mask fraction (Figure 16b, correlation −0.31) and the number of Cimel AOD observations in that day (Figure 16c, correlation 0.60). Figure 16 therefore shows that AODs and the associated cloud together explain most of the 400–650 W m−2 day-to-day variations and that cloud explains much of the ~40 W m−2 about the straight-line fit to AODs shown in Figure 16a (Figures 16b and 16c in gray).
 Principal component analysis [Preisendorfer, 1988] was performed on the data shown in Figure 16, with each component (downward shortwave, AOD, SEVIRI cloud fraction, and number of Cimel observations) normalized by its variance. The first normalized empirical orthogonal function (EOF) is 0.53, −0.50, −0.51, and 0.47 (for the four variables listed in order above) and explains 86% of the variance in the data. This is consistent with the strong (anti)correlations between downward shortwave, AOD, SEVIRI cloud fraction, and the number of Cimel observations (high AODs associated with high cloud amounts giving less shortwave) explaining most of the variability in the data (Figures 16a–16c). The second normalized EOF is 0.23, −0.58, 0.09, and −0.78 and explains 9.7% of the variance. This second EOF allows variability from the primary mode of variability expressed by EOF1 by including anticorrelations between AOD and cloud amount (lower AODs associated with fewer Cimel observations), rather than the normal correlation. This principal components analysis is therefore consistent with the simple correlations discussed above. In conclusion, disentangling the effects of dust and cloud on the surface energy budget at BBM from the data alone is challenging, due to the correlations between moisture, cloud, and dust; limits to remote sensing of cloud; and limited data.
 Before 13 June, temperatures ranged diurnally from around 25°C to 45°C. After 13 June, daily temperatures ranged from around 30°C to 45°C (with a minimum of 27°C in a haboob on 17 June). The lower nighttime minima before 13 June are consistent with the lower values of downward longwave irradiance before this time. This is consistent with the drier boundary layer, with less cloud and less dust: section 3.6.1 and Figure 8a show that the diurnal cycle in downward longwave is in fact largely controlled by cloud. Variations in downward longwave are, however, small compared with the downward shortwave, which, as noted, largely controls the variability in the net radiation and hence the sensible heating (see also variance in diurnal cycles in Figure 8a).
3.6.1 The Diurnal Cycle in the Surface Energy Balance
 The mean diurnal cycle is strongly driven by the diurnal cycle in the surface energy balance; the mean maximum net radiation is around 430 W m−2, with typically −30 W m−2 at night. Figure 8 shows that upward shortwave irradiance is approximately 33% of the downward shortwave. There is a net longwave cooling of ~50 to 150 W m−2. The phase of the upward longwave leads the air temperature, as soil skin temperature leads the air temperature (not shown). The diurnal cycle in downward longwave has a maximum at 19 UTC and a minimum at 06 UTC, showing that it is dominated by cloud (crosses in plot), which have a similar diurnal cycle (dust AODs show a weak diurnal cycle with a maximum at 13 UTC).
 The day-to-day variations in downward shortwave dominate the variations in net radiation (consistent with Figure 5). Downward shortwave is decreased by both clouds and dust. Figure 16d shows the mean diurnal cycle in the deficit of observed shortwave down from an idealized sine curve centered on solar noon. The deficit is asymmetric, consistent with the greater cloud fraction in the afternoon leading to approximately 50 W m−2 less irradiance in the afternoon compared with the morning. Together with Figure 16, this shows the importance of clouds and dust in regulating heating within the SHL and the need for prognostic clouds and dust in models. It should be noted, however, that while dust shades the surface it also heats the atmosphere [Ryder et al., 2013], while clouds largely simply scatter radiation to space.
 There has been a shortage of observations for understanding the climate of the Sahara and evaluating weather and climate models. The Fennec project has provided a comprehensive data set for this region, with the observations made during June 2011 by the Algerian ONM at Fennec supersite-1 (BBM), close to the SHL center and the summertime dust maximum, a key part of that data set. The Fennec supersite-1 data include the first observations of the diurnal cycle of the boundary layer and surface fluxes in the central Sahara, with the first ground-based lidar profiles and Sun photometer, nephelometer, reflectometer, dust sampler, and 1 Hz surface meteorology and 20 Hz winds for evaluating satellite retrievals, dust properties, and dust uplift and transport mechanisms. The Fennec automatic weather station network [Hobby et al., 2013] provides a complementary data set with a longer time duration, but no vertical profiles or observations of dust. The Fennec supersite-2 at Zouerat in Mauritania (Todd et al., submitted manuscript) is much more maritime-influenced with much less active dust uplift and lower dust loads.
 The influence of the West African Monsoon gradually increased at BBM throughout June. Most Fennec data are from an initial dry period at BBM (8–12 June) and then a period of repeated monsoon events (13–30 June). The mean diurnal cycle shows the nocturnal advection of monsoon moisture to BBM at low levels (similar to Parker et al.  for the Sahel). Cold pools were observed within these monsoon events and were often observed to be a component of the moisture transport. Overall 34% of the surface–750 hPa boundary layer moistening occurred during cold pool periods; this is expected since cold pools are found in the moister monsoon air but may also reflect the role of cold pools in moistening and cooling the SHL (see Garcia-Carreras et al., 2013). In contrast, much less cold pool outflow activity was observed at Fennec supersite-2 in June 2011. The Fennec airborne campaign observed aged cold pools from the Sahel and Atlas mountains [Ryder et al., 2013], but its distance from the area of most active moist convection (it was based approximately 700 km from SS2 and 1700 km from SS1, Figure 1) meant that no observations were made in large active haboobs.
 The day-to-day variability in net radiative heating at the surface and consequent sensible heating of the atmosphere is dominated by the variability in the downward shortwave radiation. The 400–650 W m−2 variations in daily mean downward solar irradiation can be explained by variations in AODs (correlation of –0.96) or clouds (correlation of around 0.8). AODs correlate with cloudiness (correlation of around 0.7), so it is the combined effect of dust and cloud that cause most day-to-day variations in net radiative heating. Clouds explain much of the ±40 W m−2 variability about a straight-line fit between AODs and downward shortwave. Fennec aircraft data and satellite observations [Stein et al., 2011] show that, during the afternoons, clouds are prevalent over quite extensive regions of the Sahara, showing that this result is not unique to BBM. The mean diurnal cycle shows the asymmetric effect of clouds on radiation about midday, with afternoon clouds decreasing irradiance by approximately 50 W m−2 compared with the morning. The diurnal cycle in downward longwave radiation is dominated by the diurnal cycle in cloud. The relative roles of dust and cloud in controlling the mean radiative heating may be different from their roles in controlling the day-to-day variations or the diurnal cycle, and for models the results show the importance of both prognostic clouds [Stein et al., 2011] and prognostic dust.
 Radiosondes from BBM confirm the observation of Messager et al.  that the SRL can persist throughout the day—this occurred from 8 to 12 June 2011 at BBM, and lidar data showed that this affects dust transport. The cold CBL under the residual layer was provided by ventilation of the SHL region from the Atlantic, generated by a midlatitude trough affecting northwest Africa, which appears to have resulted in an eastward displacement of the SHL. A diurnal cycle was observed in this persistent residual layer, which would be consistent with mixing over neighboring hotter surfaces [Huang et al., 2010]. However, instrument artifacts could also create such a pattern. The maximum depth of the well-mixed boundary layer was 490 hPa (6 km), with deep well-mixed boundary layers generally observed at 15 and 18 UTC, but not at 12 UTC. Very weakly stable residual layers were often observed above the growing boundary layer.
 LLJs were regularly observed: the combination of radiosondes, sodar, lidar, and surface data provides a valuable data set to evaluate models, which often give a limited representation of LLJs [Todd et al., 2008]. The strongest LLJs were all observed in the period from 13 to 30 June when strong pressure gradients close to the ITD accelerated nocturnal winds [Knippertz, 2008], with three of the four strongest LLJs being moist southerly events rather than dry Harmattan winds. The corresponding observations from Fennec supersite-2 also regularly showed LLJs, but there they interacted with daily ventilation from the coast and did not produce dust uplift frequently. The Fennec airborne campaign sampled several active LLJ and aged uplift events [Ryder et al., 2013].
 A “Saharan upper boundary layer jet” was detected in the mean diurnal cycle from radiosondes. Winds at around 650 hPa feel the effect of surface drag via the deep Saharan CBL from around 15 to 18 UTC. These winds were then observed to accelerate overnight, with the diurnal cycle in winds consistent with an inertial oscillation. This led on average to a drying at these altitudes by 09 UTC, which may contribute to the cloud minimum observed at around 06 UTC. During the day, boundary layer convection then mixes moisture upward from lower levels, which are moistened overnight by advection.
 For winds strong enough to generate dust uplift, 10 m winds were shown to provide a strong control on friction velocity. During uplift events observed dust loads were observed to depend on “uplift potentials” (the wind-based component of parameterized dust uplift, section 2.1 and Table 2), and uplift potentials are used to provide an estimation of the relative contribution of different meteorological mechanisms to the total dust uplift. The threshold used has a small effect on these contributions, with higher thresholds giving more uplift in cold pool outflows (Table 2).
 LLJs in the strong pressure gradients close to the ITD and haboobs embedded in the monsoon flow dominated dust uplift, with most dust in the moist periods. Around 54% of uplift potential and 50% of the scattering occurred at night (Table 2), almost all from cold pool outflows rather than the monsoon flow. It may be that the strong nocturnal inversions in the central Sahara inhibit mixing of nocturnal monsoon surge winds to the surface compared with moister regions farther south [Knippertz and Todd, 2012], where this has been observed previously [BouKaram et al., 2008] (at 18°N). Passive remote sensing of dust uplifted at nighttime is problematic, and similarly, use of infrared observations to detect dust in moist air can be challenging [Brindley et al., 2012]. So even without the high nocturnal cloud fractions observed, these two effects will likely bias dust climatologies which make use of infrared measurements.
 The periods of convective outflows dominated the overall dust uplift potentials (around 50% of the total). The largest events were extensive outflows visible from space, but small microburst-type events were also observed. Around 30% of uplift potential occurred between 6 and 12 UTC, consistent with the LLJ (with minimal overlap with convective outflows). Around 10% occurred from 12 to 16 UTC (and not in convective outflows), but half to one third of this occurred in 1 day of high synoptic scale winds (29 June). Haboobs also generated the largest AODs, and the two largest haboobs each generated 17% and 10% of the total uplift potential. The proximity of BBM to the maximum dustiness seen by OMI suggests that meteorology of the dust uplift observed at BBM is important for the total dust export from North Africa, although the mountains that initiate convection upstream of BBM may make haboobs more important here than in other locations. The data demonstrate for the first time from multiday in situ observations that haboobs are a key dust-generating mechanism and therefore support the hypothesis [Marsham et al., 2008a] that haboobs make a substantial contribution to the seasonal cycle of dust in West Africa. Efforts are now required to improve the very poor representation of these features in numerical models [Reinfried et al., 2009; Marsham et al., 2011a].
 These findings are briefly summarized in Figure 17. Future studies within Fennec will address each of these subjects in more detail and should provide new insights into the meteorology and climate of the Sahara, with important implications for our understanding of the global climate system.
 Fennec was funded by a NERC consortium grant (NE/G017166/1).We would like to thank Benyakoub Abderrahmane, Mohammmed Limam, and Diali Sidali (ONM) for their assistance with the supersite and indeed all at ONM Algeria for their patience and hospitality during Fennec. We would like to thank the AERONET PHOTONS team for their assistance with the Cimel Sun photometer. Ralph Burton (NCAS, University of Leeds) computed the EOFs discussed in relation to Figure 6. False color SEVIRI imagery is taken from the Imperial College Fennec Web site (http://www.fennec.imperial.ac.uk). NCEP CPC rainfall anomaly plots (not shown) were provided by Elizabeth Good (Met Office). Acknowledgment is made to the FGAM (Facility for Ground-Based Atmospheric Measurement), NCAS (National Centre for Atmospheric Science) for the use of the sodar, lidar, and radiosonde units. OMI data used in this study were produced with the Giovanni online data system, developed and maintained by the NASA GES DISC. ECMWF Operational Analysis data were provided by the NCAS British Atmospheric Data Centre (BADC, http://badc.nerc.ac.uk/view/badc.nerc.ac.uk__ATOM__dataent_ECMWF-OP). Radiosonde data were provided by the UK Meteorological Office, via the NCAS BADC (http://badc.nerc.ac.uk/view/badc.nerc.ac.uk__ATOM__dataent_GLOBRADS). We would like to thank one anonymous reviewer and Dr. Cyrille Flamant, as well as the Associate Editor, whose comments significantly improved the clarity of the paper.