Influence of Orography Upon Summertime Low-Level Jet Dust Emission in the Central and Western Sahara

Low-level jets (LLJs) drive frequent emission of mineral dust in the central and western Sahara in boreal summer. A major hotspot for this process is central Algeria, northern Mali and Mauritania, through which blow the dry near-surface northeasterly Harmattan winds, with a peak in dust emission around the low-lying Tidihelt region. North African orography is thought to contribute to the strength of the LLJ over the Bodele dust source in Chad, but its influence on erosivity over summertime source regions remains unquantified. In this paper, the contribution of central Saharan orography to the strength of Harmattan LLJs and associated dust emission frequency is tested. An idealized simulation with flattened Hoggar mountains is compared with a control using the Met Office Unified Model at 12 km horizontal resolution. In the absence of the Hoggar mountains, dust emission frequency estimated using an empirical relationship with surface wind speeds is found to decline across the entire northeasterly “LLJ alley,” including by 31% in the Tidihelt where composited jet surface winds drop from 9.0 to 7.3 m s −1 under a more easterly regime. The mountains are linked to a low-level leeward geopotential height perturbation, with a northern limb reinforcing northeasterlies through the Tidihelt. Dome-shaped elevated heating situated over the Hoggar mountains explains the difference between the simulated wind fields in the two experiments. These findings suggest that central Saharan orography plays an important role in sustaining erosive dusty conditions during boreal summer.

A robust feature of both model and satellite analyses of Saharan dust is a prominent hotspot of dust presence and emission around central Algeria (Ashpole & Washington, 2013;Evan et al., 2015;Heinold et al., 2013;Schepanski et al., 2007;Todd & Cavazos-Guerra, 2016;Wu et al., 2020), as well as more generally within the broad northeasterly Harmattan 'alley' of the northern Sahara. The central Algerian surface wind speed maximum, referred to as the Tidihelt jet in this paper, is evident in reanalysis (see e.g., Figure 13g-13i in Caton Harrison et al. [2019]) and surface observations (Chellali et al., 2011;Messaoudi et al., 2019). The Tidihelt region is unique in that unlike almost all other dust sources, it is predominantly activated by LLJs (Caton Harrison et al., 2019). A map of the Tidihelt dust source showing potential erodible salt flats is given in Figure 15 of Caton Harrison et al. (2019). The LLJ alley containing the Tidihelt jet is bounded by the Atlas mountains to the north and the Hoggar mountains to the south (see Figure 3a).

Orographic Effects on Dust-Emitting Conditions in the Sahara
Although the proximity of many Saharan dust sources to orographic channels has been noted (Evan et al., 2015;Fiedler et al., 2013;Schepanski et al., 2007), few studies have drawn an explicit link between Saharan orography and the frequency or strength of dust emission, especially for western sources such as the Tidihelt. The Bodele in northern Chad is frequently activated when a ridging Libyan high pressure system drives statically stable northeasterlies through a gap between the Tibesti and Ennedi massifs, splitting the flow upstream and accelerating winds in the exit region by up to 40% (Todd et al., 2008;Washington et al., 2006). A secondary effect from elevated heating and cooling over the mountains is also apparent. The Bodele is much larger and more active than western Saharan dust sources, but serves nonetheless as a useful analogue.
Western Saharan orography stretching from the Atlas mountains in the north to the Aïr mountains in the south has been linked to atmospheric phenomena relevant to dust emission. Kelvin waves regularly propagate along the barrier of the Atlas mountains, organizing downslope winds and jet adjustment processes responsible for severe dust storms in the Harmattan channel (Pokharel & Kaplan, 2019). A relationship may also exist between the Atlas range and the remote Saharan boundary layer, with ascent over the mountains and compensating descent over the central Sahara (Flamant et al., 2007). There is also a plausible but untested connection between orography and triggering of convection in the southern Sahara during the late monsoon season, which in turn produces convective downdrafts responsible for dust emission (Caton Harrison et al., 2021).
The Hoggar mountains south of the Tidihelt have a peak of 2,918 m, span 800 km and consist of rock pinnacles and volcanogenic formations, with widespread dust sources in their western lee and southern flank formed from outwash fluvial and paleo-lacustrine deposits (Ashpole & Washington, 2013;Prospero et al., 2002). These mountains are thought to have some role in the onset of the West African Monsoon due to their interaction with the SHL via lee cyclogenesis. It has been shown that background easterlies and northeasterlies around the mountains could be capable of initiating a leeward trough or depression as the subtropical anticyclone is amplified over the 10.1029/2021JD035025 3 of 24 elevated terrain, in turn deepening the SHL (Drobinski et al., 2005;Semazzi & Sun, 1997). The mechanics of such a process have not been demonstrated in detail, however, and have not been linked directly to dust emission. Birch et al. (2012) previously flattened the Hoggar mountains in a set of idealized model experiments using the Met Office Unified Model, showing how orography heats and deepens the convective boundary layer over the Sahara several hundred kilometers west of the main peaks, and redirects cooler maritime easterlies upstream. These thermodynamic effects could offer an additional or alternative explanation for the interaction between the Hoggar mountains and summertime dust emission, but the effect of orography upon dust emission frequency from nearby dust sources has not been explored.

Aims
In this paper, we test the effect of the Hoggar mountains upon a prominent climatological peak in summertime wind speeds over central Algeria associated with a frequent LLJ, identified here as the Tidihelt jet. Boreal summer is selected for analysis as dust emission from the central and western Sahara, including central Algeria, is high at this time of year Kok et al., 2021;Ridley et al., 2012;Schepanski et al., 2012) and because automated satellite detection of LLJ has been developed for the summer months (Caton Harrison et al., 2019). The study also estimates the impact of these winds on dust emission frequency. To achieve this, we identify the synoptic conditions favoring the Tidihelt jet, test for elevated heating and lee cyclogenesis associated with the elevated terrain and estimate the wider impacts of the orography upon dust emission frequency in the northern Saharan LLJ alley. In summary, the research aims to: 1. Quantify the effect of the Hoggar mountains upon dust-emitting winds in central Algeria 2. Identify a mechanism linking the Hoggar mountains and elevated surface wind speeds 3. Estimate the effect of the Hoggar mountains upon dust emission frequency within the LLJ alley of the northern Sahara Model and observation datasets are described in Section 2, as well as a composite method. The synoptic conditions associated with dust emission in central Algeria are described in Section 3, with the results of the model experiment presented in Section 4.

Model Data
This study uses a regional climate model (HadREM3-GA7.05) in a limited area configuration of the GA7.05 Unified Model (UM) (Walters et al., 2019), closely based on the model configuration used in UKCP18 climate projections (Murphy et al., 2018). It is set up in atmosphere-only mode with a limited area domain centered over North Africa on a 12 km horizontal grid with 57 vertical levels. The UM dynamical core solves deep-atmosphere non-hydrostatic equations with a semi-implicit, semi-Lagrangian formulation discretized onto a regular latitude-longitude grid with terrain-following hybrid height coordinates (Walters et al., 2019). In this configuration, convection is parameterized and sea surface temperatures and sea ice extents are prescribed using the analyses of Reynolds et al. (2002), in addition to aerosol properties and cloud droplet number concentration derived from the MACv2-SP dataset for the historical (Stevens et al., 2017) scenario. The method used to implement these aerosol effects is identical to that used in the 12 km RCM in UKCP, although with a difference source dataset. As dust emission is not explicitly simulated, the effects of winds upon dust emission are derived empirically based on satellite-observed dust source LLJ activity, described in Section 2.2.
Of particular relevance to the simulation of LLJs is the boundary layer scheme. While inertial oscillation is handled by the dynamical core, the UM boundary layer scheme (Brown et al., 2008;Lock et al., 2000) is responsible for all vertical mixing by turbulent motions and is therefore critical for LLJ decoupling and decay. In an unstable boundary layer, diffusion coefficients (K profiles) are defined for sources of turbulence from both the surface and cloud top, whereas in a stable regime a local Richardson number scheme (Smith, 1990) applies. Shallow and deep sub-grid cumulus convection is handled by the UM mass-flux convection scheme, with an extension to include parameterization of downdrafts (Gregory & Rowntree, 1990;Walters et al., 2019).
Lateral boundary conditions and initial conditions are supplied to the limited area model by the ERA-Interim reanalysis (Dee et al., 2011) (Allen & Washington, 2014), it is capable of realistically reproducing wind variability associated with synoptic and seasonal variations (Roberts et al., 2017) which is important for lateral boundary conditions.

Observations
A key characteristic of the summertime Sahara is that the majority of dust sources, including many in the Harmattan channel, are regularly influenced by cold pool outflows (CPOs) (Caton Harrison et al., 2019). In this paper, data from the Spinning Enhanced Visible and Infrared Imager (SEVIRI) are used to identify days with dust present and to compare simulated wind speeds with dust emission frequency. Dust is identified in SEVIRI from Brightness Temperatures at 15 min intervals and 3 km nadir resolution using the SEVIRI dust flags approach of Ashpole and Washington (2012) and subsequently filtered to identify plumes associated with LLJ activity following the method in Caton Harrison et al. (2019). Critically, this allows dust associated with LLJs to be distinguished from CPO dust. Although SEVIRI is commonly used for dust detection (Ackerman, 1997;Lensky & Rosenfeld, 2008;Schepanski et al., 2007), split-window techniques using wavelengths from 8 to 12 μm such as SEVIRI dust flags are also sensitive to column water vapor and cloud . Central Algeria is partly spared from such drawbacks as it is exceptionally dry and has a high percentage of cloud-free days (e.g., see Figure 13 in Caton Harrison et al. [2019]).
Radiosonde data used in this paper to identify synoptic conditions linked to dust emission are obtained for In Salah (27.19°N,2.47°E), approximately 150 km upwind (northeast) of the Tidihelt dust sources. Daily soundings at 12:00 UTC are sourced from the Integrated Global Radiosonde Archive (Durre et al., 2006) maintained by the National Centers for Environmental Information.

Model Validation
The UM is selected as the modeling system for this project as it has demonstrable fidelity in simulating both the large-scale synoptics and the jet core wind speeds within the boundary layer. A 12 km horizontal grid-spacing configuration of the UM has been deployed (as 'Africa-LAM') to support the Fennec Campaign of in-situ observations in 2011 and 2012 . Comparisons between Africa-LAM and observations from supersite 1 located at Bordj-Badji Mokhtar in southern Algeria (approximately 500 km south of the Tidihelt region defined in Section 2.5) reveal 'excellent agreement' between model wind speed profiles at 06:00 UTC and Lidar measurements for Harmattan LLJs (Allen & Washington, 2014), equivalent to the northeasterly jets considered in this paper. Allen et al. (2015) also find a hit rate of 85% in simulating morning LLJ winds at the surface compared to Fennec automatic weather station data, but they note that the frequency of the highest ( 10 m s −1 ) winds is underestimated. Dropsonde measurements from Fennec flights show that forecasts accurately represent large-scale wind fields around the SHL but underestimate morning wind speeds (Engelstaedter et al., 2015). Figure 1 compares the 12 km configuration of the UM used in this paper with the approximately 31 km resolution ERA5 reanalysis. With a severe paucity of in-situ observations in the central Sahara, there is little for reanalyses to assimilate and dust extinction is not generally included in temperature or humidity retrievals from satellite infrared sensors, leading to biases in these assimilated variables (Weaver et al., 2003). As a result, significant disagreements exist between reanalyses over North Africa. This has led to uncertainty in their ability to faithfully represent the SHL , mesoscale convective systems (Roberts & Knippertz, 2014) and the intertropical discontinuity (Roberts et al., 2015). Comparison against ground stations from other parts of the world indicates ERA5 outperforms other reanalyses in representing diurnal variability of surface winds (Ramon et al., 2019), but an extensive evaluation of ERA5 winds against ground stations and other reanalyses has not yet been carried out over the Sahara. We adopt it here only for a broad comparison of synoptic features, given that it likely inherits underestimation of peak Saharan surface winds shown in ERA-Interim (Allen & A close match between the large-scale wind field is evident in comparing the UM and ERA5 (Figures 1a and 1b). Northeasterlies enter the region of the SHL via a gap between the Atlas and Hoggar mountain ranges, as well as Atlantic inflow from the northwest. A minimum in wind speeds occurs at 22°N within the climatological ITD. The highest winds are found along the Atlantic coast where a persistent coastal LLJ linked to the Azores High and SHL overlies the Canary Current (Soares et al., 2019). Crucially for the purposes of this project, the UM accurately reproduces a wind speed maximum in central Algeria, evident in both the reanalysis and surface observations (Chellali et al., 2011). Wind speeds in the core of this maximum are 1-1.5 m s −1 higher in the UM compared to ERA5 (Figure 1e), which may be due to a deeper SHL (Figures 1c and 1d), driving a sharper pressure gradient through central Algeria ( Figure 1f) as well as due to the higher spatial resolution of the UM (12 km compared to 30 km in ERA5).
An interesting feature of the Tidihelt jet is that it appears as a small local wind maximum year-round (Figure 2), including when the SHL is situated much further south than its central Saharan position in JJA (Lavaysse et al., 2009). This shows that a local control is important for the position of these high winds, rather than simply being the result of synoptic pressure patterns. Nonetheless, the jet is at its strongest in JJA ( Figure 1b). Whereas in DJF, MAM, and SON the top of the jet feature is situated at 850-800 hPa, during JJA a much deeper boundary layer appears to mix momentum from the jet to greater heights of 750-700 hPa (not shown).

Orography Experiments
Two simulations are run. In the first (CONTROL, Figure 3a), the UM is run with full orography over the limited area domain. In the second run (FLATHOGGAR, Figure 3b), the region encompassing the Hoggar mountains is smoothed to a uniform level. Unlike Birch et al. (2012), we do not remove the Aïr mountains at 19°N, 9°E as our goal was to isolate the role that mountains adjacent to the Tidihelt region have upon winds there; furthermore, the Aïr Mountains are an important site of late summer convective triggering (see Redl et al. [2015], Figure 6 therein) which could have an impact upon the low-level monsoon circulation. The same lateral boundary conditions are supplied by ERA-Interim in both experiments, meaning that although the removal of the Hoggar mountains may influence the large-scale flow beyond the limited area domain, this effect is not accounted for in this paper.
One approach tested for flattening the Hoggar region was to apply a mask and level all terrain therein to a maximum of 350 m above sea level. This introduced sharp horizontal discontinuities, especially toward the east of the domain around 12-14°E which produced local wind acceleration in the simulations. The solution adopted here is to smooth the mountains by cloning terrain from the lowlands in the central Saharan Erg Chech and El Djouf (4°W-0°E) over the Hoggar plateau, introducing a more gradual incline toward the higher elevation of eastern Algeria and Libya. Terrain from the western Sahara is selected for cloning as it is homogeneous in elevation over spatial scales of hundreds of kilometers and therefore best approximates a flat sand sheet with no orography.
Edits are applied to a 1 km resolution topography dataset prior to generating the UM ancillary files describing the mean and sub-grid properties of the orography. The effects of flow blocking, orographic form drag and orographic gravity wave drag are then calculated consistently following the method described in Walters et al. (2019) (see Sections 2.6 and 3.5 therein). Orographic effects at the smallest scales (at which buoyancy effects are negligible) are represented by an effective roughness length (i.e., indirectly) while sub-grid orographic effects at larger scales up to the model resolution are represented with a drag scheme originally developed by Lott and Miller (1997) with slight modifications described in Section 3.5 of Walters et al. (2017). No changes are made to albedo, which is tested in more detail in Birch et al. (2012) and shown to exert a smaller influence on the boundary layer than the orography itself. Surface roughness is also not edited in these experiments beyond the effective roughness length described above.

Composite Method
The composite method in this paper selects high northeasterly surface wind days with visible dust emission.
Compositing is used in this analysis rather than averaging over the whole simulation for two reasons. Firstly, although northeasterlies are the most common wind regime in the Tidihelt region, other wind regimes are active through summer due to occasional incursions from monsoon winds and Atlantic inflow. Secondly, high wind days are disproportionately important for dust emission. The bulk of the results presented in this paper are obtained from a comparison of composited days between the two experiments. An analysis of the overall effect on the potential for dust emission across all days is presented in Section 4.4, however.
A sub-domain within the limited area of the model simulations is delineated and corresponds to the Tidihelt Depression in central Algeria, described in more detail in Caton Harrison et al. (2019). The boundaries are 25-28°N, 1.5°W to 3°E (the box region in Figure 5). Henceforth, this is referred to as the TID region.
Mornings with high surface winds and visible local LLJ-linked dust emission in the control experiment TID region are selected for compositing. Dust emission is not computed in the model simulations, hence information about emission must be obtained from the contemporaneous satellite record. The period of June, July and August (JJA) 2004-2007 is used as this contains an overlap between the period for which UM data is available (1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007) and the period for which SEVIRI dust tracking has been performed (2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017). Each selected day must fulfill the following conditions for the day to count as a dust day: 1. At least 500 total dust pixels visible between 05:00 and 13:00 UTC in SEVIRI in the TID region which the algorithm from Caton Harrison et al. (2019) identifies as LLJ dust 2. Any dust observed is freshly emitted (within the same morning) 3. Mean control Experiment 10 m wind speed within the TID region exceeds 7 m s −1 4. Mean control Experiment 10 m wind direction within the TID region for the day is between 0° and 90°A wind direction constraint is included in order to limit the analysis to a northeasterly wind regime associated with the prevailing dry Harmattan. As identified in Section 2.3, the existing analysis of UM performance in Allen and Washington (2014) indicates that fidelity is excellent for northeasterly jets. On the other hand, it is relatively poor for southwesterly jets embedded within the monsoon.
The 30 days with the highest mean wind speed and which also fulfill the conditions above are selected for inclusion within the composites. Composites are averaged over 05:00-12:00 UTC to target a period in which both the core strength of the jet peaks and the mixing of momentum to the surface occurs. Note that an extra hour is included on the end of the SEVIRI dust observation window as occasionally dust plumes emitted in the mid-morning do not become visible until around midday.
The 7 m s −1 wind speed threshold chosen here is conservative compared to estimated dust emission thresholds at surface stations in central and southern Algeria (Cowie et al., 2014), but provides a good sample of days to choose from given the limited time analysis time period of four summers; 46 days fulfilling each composite condition are found, of which the top 30 are selected. For a clearer picture of the likely dust emission threshold for the TID region, an analysis of the relationship between simulated winds on these composite days and observed dust is given in Section 4.4.

Synoptic Conditions
Lag-lead composites of 12:00 UTC soundings from In Salah (Figure 4) reveal that a cold northeasterly boundary layer anomaly accompanies strong winds and dust emission over the Tidihelt region. Cold anomalies of −0.5 to −3 K, significant at the 95% level, are observed up to 700 hPa through from Day −3 to Day +3. By Day −1 there is a significant northeasterly anomaly of 4.8 m s −1 at 925 hPa, backing to northerly at 700 hPa. The coldest layer is found at 925-850 hPa on Day 0 along with a wind anomaly of 5.6 m s −1 at 925 hPa and a 3.6 m s −1 anomaly observed at the surface level of 982 hPa. Days +1 to +3 see a continued cold anomaly with winds backing more northerly.
Patterns in the fields from the UM control experiment map on well to observations from Figure 4 and reveal the synoptic processes driving the wind anomalies (Figures 5a-5d). A cold anomaly extends from the Mediterranean to central Algeria, advected by northeasterlies which reach their westernmost extent on Days 0 and +1 (Figures 5c and 5d). This strengthens winds through the Harmattan LLJ alley into the core of the SHL and also drives an intense warm anomaly on the Atlantic coast by blocking cooler maritime inflow. Intensification of subtropical low-level high pressure over the eastern Sahara associated with intrusion of a midlatitude trough and cold inflow from the Mediterranean has long been linked with strengthened Harmattan winds (Kalu, 1979, Figures 5.4 and 5.5 therein). Synoptic conditions shown in Figure 5 resemble a typical Mediterranean cold surge event which can persist over North Africa for up to 10 days alongside an enhanced upper tropospheric ridge-trough pattern (Vizy & Cook, 2009). Upper level ridging over the Atlas mountains in the UM simulation is accompanied by a trough over the Mediterranean (Figures 5e-5h) which propagates eastward through the composites. At low levels, a low pressure anomaly is present south of 28°N and east of 0°E which shifts westward from day −2 (e) to day +1 (h). North of 28°N, anomalous high pressure develops and intensifies to a maximum at day 0 (g) over and south of the Atlas mountains, producing conditions associated with a high North African Dipole Index (NAFDI) (Rodríguez et al., 2015).
In the remainder of the paper, we study the low-level circulation patterns for the lag 0 composite days in both the control and FLATHOGGAR simulations.

Wind Speed and Vertical Structure
925 hPa composite winds in the control experiment ( Figure 6a) exhibit a maximum northwest of the Hoggar mountains, extending from 6°E to the Algerian border with Mali and Mauritania. Northeasterlies between 56° and 79° account for over 40% of composite winds averaged across the TID domain (Figure 6b). TID is also situated at a confluence with southeasterlies along the western flank of the Hoggar mountains. The northern limb of the SHL circulation is found around northern Mali at 24°N, with northerlies across Mauritania.
A substantial reduction in wind speeds occurs over the Tidihelt region in the absence of the Hoggar mountains (Figure 6c), with mean composite winds within TID dropping from 9.0 to 7.3 m s −1 . A slight local maximum nonetheless persists over central Algeria. A shift in the wind regime can be seen over the TID region, with winds in the 56-79° range reduced (Figures 6b and 6d) in favor of easterlies (79-101° on the wind rose), which also prevail over the region previously occupied by the Hoggar mountains. Southeasterlies are no longer evident in southern Algeria and the SHL circulation appears to be shifted southward, with a northern limb extending down to 22°N and reduced northerlies in eastern Mauritania (Figure 6c).  It should be noted that a potential drawback of a compositing approach here is that data in the control experiment is selected for high wind speeds, whereas data in the counterfactual FLATHOGGAR experiment is not; removal of orography could affect the timing of high wind events and the wind regime in which they occur and hence cause winds to peak on different days. When the same analysis as in Figure 6 is carried out for the entire dataset instead of composite days, a similar result emerges in a subtler form; the dominant shift is still a reduction of wind speeds proximate to the Tidihelt (by 10.1% instead of the 18.9% obtained with compositing) and a slight preference for easterlies over northeasterlies (not shown). The effect of orography upon modeled winds is therefore greater than usual on days with synoptic forcing driving an intense northeasterly jet. There is no indication, however, that the strongest winds seen on composited control days occur at other times in the FLATHOGGAR results. Further evidence of this may be found in Section 4.4 (Figure 13a) in which non-composited wind speed distributions are compared for the two experiments. A composite cross section across the TID region in the control experiment reveals that the Tidihelt jet is a low-level nocturnal feature, extending to approximately 1,800 m above sea level at 00 UTC (Figure 7a). The jet core is close to the surface at 1,000 m above sea level or approximately 100 m above the surface. During the day, winds within the boundary layer are well-mixed with a much slower core below 10 m s −1 extending up to above 2,000 m above sea level (Figure 7b). Although the nocturnal jet extends from 1 to 6°E, the highest winds are found south of the Tademaït Plateau in the low-lying Tidihelt Depression, corresponding to the TID region (25-28°N, 1.5°W to 3°E). In the FLATHOGGAR experiment (Figures 7c and 7d), the jet core is significantly eroded and less clearly defined, although the winds are still accelerated nocturnally (Figure 7c).

Leeward Circulation
The difference between the two experiments at 850 hPa reveals a circulation pattern centered over the TID (Figure 8). The circulation has a northern limb of northeasterly anomalies through central Algeria and a southern limb of southeasterlies and easterlies in the lee and over the Hoggar mountains (Figure 8b). This is associated with a geopotential height perturbation of −15 to −20 m over the TID region at 850 hPa ( Figure 8a), with significant differences of 5-10 m as far west as the Atlantic coast.
East-west cross sections through the Hoggar mountains indicate a diurnal variation in the circulation pattern. In both the control (Figures 9a-9d) and FLATHOGGAR (Figures 9e-9h) experiments, easterlies persist nocturnally, albeit weakened on the western slopes in the control. During the day, however (Figures 9c and 9d), easterlies drop to zero between 2° and 5°E in the control, whereas in FLATHOGGAR they are maintained at 3-5 m s −1 (Figures 9g and 9h). This suggests an upslope wind in the control experiment driven by enhanced sensible heat fluxes over the elevated terrain driving a thermally direct circulation (Wolyn & Mckee, 1994;Zängl & Chico, 2006). In this instance, however, no actual upslope winds develop as the background flow is too strong. Differences between the meridional circulation are more stark; whereas in FLATHOGGAR a mild southerly component is present at 06:00 UTC (Figure 10f) and eroded entirely during the day (Figure 10g), in the control experiment a southerly flow develops overnight over the peak of the Hoggar and reaches a maximum of up to 5 m s −1 at 06:00 UTC (Figure 10b) at which point it is located over the western mountain lee slopes. The result is a net southeasterly into the Tidihelt Depression. During the day, this flow weakens and shifts further westward (Figure 10c).
The Hoggar mountains are subject to intense heating during the day. The difference between potential temperature fields of the two experiments becomes greatest at 12:00 UTC (Figure 11c), with a peak over the summit extending up to 3,000 m in a dome shape. During the night, this heating perturbation shifts westward (Figure 11a, 11b, and 11d) and is situated within a residual layer above the surface over 0-5°E. This elevated heating due to the presence of the Hoggar mountains induces a horizontal temperature gradient over the plateau (about 1,500 m elevation) relative to the low lying Sahara to the west (about 300 m elevation). A gradient between the cooler maritime west and the SHL exists in the absence of the mountains (not shown) but is greatly intensified when elevated terrain is present. The existence of elevated heating and baroclinicity suggests that the Hoggar mountains may be capable of setting up a thermal low or trough within the much broader SHL, deepening and extending it.
The Tidihelt jet is a nocturnal phenomenon; 925 hPa composite winds in the Tidihelt region reach their maximum at midnight UTC and minimum at 13:00 UTC, whereas the minimum mean 925 hPa geopotential height in the lee of the Hoggar mountains occurs at 17:00 UTC (not shown). This temporally shifted relationship aligns with the theoretical behavior of heat lows which develop to their lowest central pressure in the afternoon and evening. Being out of quasi-geostrophic balance, however, the peak winds and relative vorticity occur several hours after the pressure minimum (Rácz & Smith, 1999;Spengler & Smith, 2008). The exact timing of the peak in low-level winds can also depend on the Coriolis parameter and its effect on inertial oscillation (Rácz & Smith, 1999); this is tested by Heinold et al. (2015) who show that a later timing of maximum supergeostrophy (i.e., closer to sunrise) is favored at 20°N compared to 30°N, with the Tidihelt at a latitude of about 26°N. A daytime minimum in low-level winds is driven by intense mixing of the convective boundary layer, which inhibits horizontal flows (Parker et al., 2005). Furthermore, it is plausible that the background easterlies act to shift the center of the heat low westward to its mean position seen in Figure 8a, similar to the effect of background easterlies on the Australian heat trough (Spengler et al., 2005).

Geostrophic Wind Analysis
To test the role that elevated heating could have upon winds in the Tidihelt notwithstanding any additional dynamic effects, we estimate the geostrophic winds induced by a heating anomaly equivalent to the difference between the control and FLATHOGGAR experiment. It should be noted that as the diabatic heating and cyclonic circulation are out of phase the real Tidihelt jet is not in geostrophic balance; furthermore its peak core wind speeds are subject to supergeostrophy due to inertial oscillation. These estimates instead serve to identify the expected spatial pattern of winds induced by thermal forcing for comparison with Figure 8.
Atmospheric thickness is estimated for the 925-700 hPa layer from the temperature profile of the control and FLATHOGGAR experiments to account for the heating effect of the Hoggar mountains. Assuming hydrostatic balance, the atmospheric thickness in a given layer corresponding to a temperature profile is given by the hypsometric equation: in which z 1 and z 2 are the lower and upper heights of the layer between pressure levels p 1 and p 2 (925 and 700 hPa in this instance), R d is the gas constant for dry air, g is gravity, and T v is the virtual temperature.
The difference between the estimated atmospheric thickness of the two experiments ( Figure 12a) shows a thickness perturbation of over 10 m centered over the lee of the Hoggar mountains, in close resemblance to the 850 hPa pattern (Figure 8a). Negative differences (warmer temperatures in the control) extend through central Algeria and into Mali, with evidence of positive differences (cooler temperatures in the control) along the west coast. This pattern is comparable to the SHL itself ( Figure 1d) but its core is situated slightly further east over the Hoggar mountains themselves. Nonetheless, these results suggest the presence of the mountains is capable of introducing elevated heating which could deepen or extend the SHL.
To understand the effect such heating could have upon the winds, we compute the geostrophic wind change induced if geopotential heights at an arbitrary level (within 925 to 700 hPa) were lowered by the thickness differences shown in Figure 12a. The results show a circulation pattern centered around the thermal low (Figure 12b), with peak geostrophic winds in excess of 5 m s −1 within and upstream of the TID domain where the pressure gradient peaks. This circulation closely resembles the spatial structure of the wind difference between the control and FLATHOGGAR experiments at 850 hPa (Figure 8b). Such a similarity suggests that heating alone is capable of explaining the pressure pattern and that a local maximum in winds over the Tidihelt is the result of a thermodynamically forced nocturnal LLJ. Our explanation for lee cyclogenesis over and west of the Hoggar mountains does not rely on the generation of an upstream anticyclone at low levels to enhance northeasterly winds as shown in the comparable idealized model experiment of Semazzi and Sun (1997). In contrast with their analysis, our composite appears to show only minor differences in anticyclonic circulation on the windward side of the Hoggar mountains (Figure 8a) although an enhanced anticyclonic return flow is observed atop the leeward circulation, especially above 500 hPa. Instead, the results suggest that the enhanced elevated heating of the boundary layer is the primary control upon the deepening of the eastern SHL and enhancement of the winds through the Tidihelt. A low level lee circulation over central and southern Algeria appears in each climatological season (not shown), suggesting it is a stationary feature which helps maintain the elevated winds in the Tidihelt through the year (Figure 2), with peak strength when the lee circulation is collocated with the SHL in summer.

Role in Dust Emission Frequency
The Tidihelt depression is an important and frequently activated dust source (Ashpole & Washington, 2013;Caton Harrison et al., 2019;Schepanski et al., 2007). Our results show that the Hoggar mountains play an important role in accelerating low level winds over the region. We hypothesize that the mountains could therefore be partly responsible for the prominence of this dust source, meaning they are important for both erosivity on their leeward flank. To test this hypothesis, an estimate of the relationship between wind speeds and dust emission is needed. To achieve this, we compare simulated wind speeds with satellite observations of LLJ dust plumes. Unlike previous results, no compositing is applied in this analysis as the aim is to estimate the net effect of orography upon dust emission conditions. Results are instead calculated from all model output from JJA 2004-2007 between the hours of 05:00 and 13:00 UTC (i.e., when freshly emitted LLJ dust is visible to satellite).  The method for distinguishing LLJ dust from CPO dust in SEVIRI imagery is described in Caton Harrison et al. (2019). In brief, the timing of emission, plume geometry and distance from deep convection is combined in a multiple regression model to assign a value from 0 to 1 to each discrete dust plume masked in SEVIRI data, with higher values indicating a higher probability of being LLJ-associated (other plumes are assumed to originate from convective CPO activity).
Here, we count dust emission events over the Tidihelt region and assign them to a wind speed bin corresponding to the mean 10 m wind speed simulated over the TID domain for the timestep at which the dust is observed. Dust counts are obtained as hourly totals and compared to hourly UM values. Only LLJ dust emitted the same day and within the TID domain is included. In addition, if the sum of cloud pixels (identified where BT 10.8 μm < 270 K, following Caton Harrison et al. [2019]) in the domain for the hours of 02:00 and 18:00 UTC exceeds 500, the day is considered cloud masked and skipped. The results of this analysis (Figure 13b, dark gray bars, with dust frequency as red crosses) give an approximation of the relationship between simulated wind speed in the control and observed LLJ dust emission across JJA 2004-2007. The most commonly occurring wind speeds are in the 3-5 m s −1 range, but the winds associated with the highest frequency of dust emission events are found in the 10-12 m s −1 range. This exponential increase in dust emission frequency at higher wind speeds is consistent with the theoretical cubic dependence of dust emission on surface wind speed (Marticorena & Bergametti, 1995). As lower wind speeds are much more common, the counts of dust events in each bin are more even (Figure 13c, dark brown) relative to the distribution of frequencies for each wind speed bin in Figure 13b. For example, although the frequency of dust emission events is quite low in the 4-5 m s −1 range (Figure 13b), the total number of emission events for this bin in the control experiment (Figure 13c) is comparable to that for the 10-11 m s −1 range as there are so many more occurrences of the former wind speed. Note that the very highest wind speed bin with any data (12-13 m s −1 ) has only 16 occurrences in the control experiment, which may explain why the dust emission frequency drops off again here.
The FLATHOGGAR experiment is observed to shift the 10 m wind speed distribution within the TID domain in favor of lower wind speeds (Figure 13b, light gray bars), with the largest differences found in the high tail; for example, wind speeds in the 8-9 m s −1 range decrease by over 50%. To estimate the effect of this upon dust emission in the Tidihelt, we assume the relationship between wind speed and dust emission holds as a constant between experiments and adjust the frequency of dust emission in each wind speed bin accordingly. For example, 2.8% of wind speed events falling into the 8-9 m s −1 wind speed bin have an LLJ emission event observable in SEVIRI at the same time (18 of 643 counts), so for the FLATHOGGAR the estimated dust counts would be 2.8% of 311 counts, or 8.7 counts. This results in a "simulated" distribution of dust emission counts for the FLATHOGGAR experiment TID region (Figure 13c, light brown bars). As high wind speeds are so important for dust emission, the effect of changes in the tails of the wind speed distribution is substantial; overall the TID region sees an average reduction in dust emission frequency of 31%.
As Figure 8b shows, the effect of the Hoggar mountains on the wind field is not unique to the Tidihelt region. Instead, wind speed changes are observed through the entire LLJ alley. To estimate the wider effect of this shift on the entire model domain, the relationship between wind speed and dust emission identified for the TID region in Figure 13b is assumed to hold across the entire domain, and an estimated FLATHOGGAR dust emission frequency distribution is calculated at each grid box based on the wind speed distribution at that grid box in FLATHOGGAR. The difference between the two experiments' estimated dust emission frequency (Figure 13a) indicates how dust emission frequency in the central and western Sahara might be different in the absence of the Hoggar mountains.
Removing the Hoggar mountains is estimated to have the greatest impact upon dust emission frequency in the Tidihelt region, with percentage decreases over 50% in the 0 to 4°E region most strongly impacted by a combination of background northeasterlies and the leeward circulation observed in Figure 8b. Effects are also observed further north close to the southern foothills of the Atlas mountains as well as over northern Mauritania. To give an indication of the position of this wind speed reduction relative to prominent dust sources, locations with known LLJ-related dust sources from the SEVIRI record are marked in Figure 13a. These are locations where a cluster of at least four SEVIRI pixels are each flagged with 10 or more LLJ dust emission events from June, July and August of 2004-2017 following the method described in Caton Harrison et al. (2019). As well as sources within central Algeria, reduced wind speeds are observed to affect dust emission frequency at sources in northern Mali, Mauritania and western Algeria. By contrast, the mountains are responsible for increased estimated emission frequencies in the immediate lee as well as upstream in western Libya (12-16°E) as a result of flow blocking and an upslope wind flow induced by the elevated heating discussed in Section 4.2, though these regions do not contain clear satellite-observed LLJ dust sources.
This counterfactual set of results comes with caveats. First and foremost, discussing the impact of the mountains in terms of an alternative situation in which those mountains did not exist implies that other conditions would be held constant in such a scenario. Orography is not only responsible for increased erosivity due to accelerated surface winds, however; a large proportion of active dust sources appear to be located close to Saharan mountains around alluvial fans, chotts and sebkhas in the foothills (Middleton & Goudie, 2001;Prospero et al., 2002;Schepanski et al., 2007), suggesting mountains drainage is also a factor in local erodibility. A second caveat is that dust emission thresholds are not constant across North Africa, and a shift in emission frequencies applicable to the Tidihelt does not necessarily hold elsewhere; for example, emission thresholds are thought to be higher in northern Algeria (Cowie et al., 2014). In addition, the lateral boundary conditions are held constant in this experiment but removal of the mountains may be expected to feed back upon the wider circulation. Lastly, as aerosols are prescribed, the modification to the atmosphere in the FLATHOGGAR experiment does not include dust radiative feedbacks, which may directly alter the local circulation via redistribution of heating within the atmospheric column (Miller & Tegen, 1998) including boundary layer stability and LLJ formation (Miller et al., 2004;Pérez et al., 2006), or indirectly, for example by introduction of ice condensation nuclei (DeMott et al., 2003;Klein et al., 2010;Price et al., 2018). The advantage of the present setup is its relative simplicity, isolating the local impact of the orography directly upon the winds as an empirically associated proxy for dust emission frequency, and the results show that even a moderate shift in the wind speed distribution can have large impacts on dust emission, but a more sophisticated setup is needed to consider the mediating effects of varied land surface conditions, feedbacks on the wider circulation and feedbacks from dust.

Summary and Conclusions
The role of the Hoggar mountains in dust-emitting wind conditions within the northeasterly LLJ alley through the central Algerian Sahara has been tested with parallel Met Office Unified Model (HadREM3-GA7.05) experiments. Lateral boundary conditions are supplied by the ERA-Interim reanalysis and a composite method is derived using dust plume observations from SEVIRI to identify days with evidence of both high northeasterly winds and visible LLJ-related dust emission within the Tidihelt region of central Algeria, chosen as a representative dust source. Reanalysis, in-situ observations and model simulations suggest a local maximum in wind speeds exists here, identified in this research as the Tidihelt jet.
High wind and dust conditions within the Tidihelt during boreal summer are associated with a long-lived northeasterly cold anomaly within the boundary layer. This is in turn linked to Mediterranean inflow from an intensified subtropical anticyclone and the passage of an enhanced upper level ridge-trough pattern, typical of cold surge conditions.
To test the role of the Hoggar mountains in the existence and strength of the Tidihelt jet, a control experiment is compared with a counterfactual experiment in which the Hoggar mountains are smoothed to be continuous with the Saharan sand sheet over the region of the SHL. The two experiments are run in parallel for the period 2004-2007 to overlap with available SEVIRI data and the 30 composite high wind, high dust days are compared.
The following results are found when the Hoggar mountains are removed in the composites: 1. Mean composite wind speeds within the Tidihelt region decline from 9.0 to 7.3 m s −1 , although a slight local maximum persists over central Algeria 2. The wind regime shifts from favoring northeasterlies to a more easterly prevailing wind 3. The Hoggar mountains are responsible for a geopotential height perturbation at 850 hPa of −15 to −20 m over the TID region, but with significant differences extending as far west as the Atlantic coast 4. The northern limb of this cyclonic circulation is collocated with the Tidihelt jet 5. A dome-shaped daytime elevated heating structure situated over the Hoggar mountains can explain the simulated wind field difference between the experiments, indicating an orographic thermal low is responsible for peak Tidihelt winds 6. Based on an empirical relationship between surface wind speeds and LLJ dust observed in SEVIRI imagery, an absence of Hoggar mountains is estimated to reduce dust emission frequency over the Tidihelt region by an average of 31% The results of this orography experiment show that the Hoggar mountains play a direct role in the surface wind field of the northern Sahara, including the dusty northeasterly LLJ alley through central Algeria. Elevated diabatic heating is shown to be an effective explanation for accelerated low-level flow around a shallow heat low disturbance centered around the lee of the Hoggar mountains. This, in turn, is estimated to have a substantial role in dust emission frequency in LLJ-dominated regions.
This work contributes to a body of literature demonstrating the role that orography plays in both the erosivity and erodibility of key western Saharan dust sources. However, it is limited in scope to LLJs, which are thought to play a secondary role in summertime dust emission. The availability of convective-permitting simulations (e.g., Heinold et al., 2013;Knippertz et al., 2009; and CPO parameterizations (e.g., Grandpeix & Lafore, 2010; Pantillon et al., 2015) means more attention could be devoted to orographic impacts upon convective triggering and cold pool formation around southern Saharan and Sahelian dust sources. The dust budget from year to year in the Sahara is a function of both the availability of erodible material and the frequency and strength of mesoscale meteorological emission mechanisms. Efforts to understand the controls on these emission mechanisms will allow model estimates of dust emission to be constrained via improvements to driving wind fields.