An analysis of the September 2015 severe dust event in the Eastern Mediterranean

In September 2015 one of the severest and unusual dust events on record occurred in the Eastern Mediterranean. Surprisingly, state-of-the-art dust transport models were unable to forecast the event. This study details the reasons for the failure of the models to forecast the dust storm and presents simulations of the event at convection permitting resolution with the modelling system ICON-ART. The results allow for an in-depth analysis of the influence of the synoptic situation, the complex interaction of multiple driving atmospheric systems and the mineral dust radiative effect on the dust event. A comparison of the 5 results with observations reveals the quality of the simulation results with respect to structure and timing of the dust transport. The forecast of the dust event is improved decisively. The event is enabled by the unusually early occurrence of an active Red Sea trough situation with an easterly axis over Mesopotamia. The connected sustained organized meso-scale convection produces multiple cold-pool outflows responsible for intense dust emissions. Complexity is added by the interaction with an intense heat low, the land inward penetrating Eastern Mediterranean sea-breeze and the widespread occurrence of super-critical 10 flow conditions and subsequent hydraulic jumps in the vicinity of the Dead Sea Rift Valley. The newly implemented mineral dust radiation interaction leads to systematically more intense and faster propagating cold-pool outflows in the runs including the mineral dust radiation interaction.


Introduction
Mineral dust aerosol plays an important role for the environment.The transport of mineral dust within the atmosphere has an effect on physical processes, chemical composition and biological systems on various temporal and spatial scales (Carslaw et al., 2010;Shao et al., 2011a;Boucher et al., 2013).Therefore, a need exists to correctly represent mineral dust and its effects in atmospheric models.
In the Eastern Mediterranean the impact of mineral dust on the environment, human population and traffic is important due to the unique environmental setting and great population density.Dust events in the Eastern Mediterranean (EM) are usually associated with strong south-westerly and southern flows in the region, although events have been reported under easterly flow conditions (Dayan et al., 1991;Levi and Rosenfeld, 1996).As a result, the most important remote dust source regions for the EM are situated in north-western Africa and the southern Arabian peninsula (Ganor, 1991;Kubilay et al., 2000).According to a subjective synoptic classification conducted by Dayan et al. (2008), the synoptic scale systems associated with dust transport wide (Knippertz, 2014).Therefore, the existence of CPOs detectable in satellite observations is suspected as the main reason for the failure of state-of-the-art dust transport models, as all of them operate at resolutions with fully parametrized convection.
In order to overcome this problem the ICON-ART (ICOsahedral Nonhydrostatic -Aerosol and Reactive Trace gases) global modelling system (Zängl et al., 2015;Rieger et al., 2015) is used in this study.It is capable of local grid refinements, in this study the finest nest has a convection permitting grid spacing of 2.5 km.The ART extension (Rieger et al., 2015) allows for on-line simulation of dust processes in ICON.The mineral dust radiation interaction has been implemented in ICON-ART as a part of this study as it has been shown to be of great importance for atmospheric processes (Pérez et al., 2006;Heinold et al., 2008;Bangert et al., 2012;Rémy et al., 2015).The mineral dust radiative effect is calculated on-line based on Mie-calculations and feeds back on atmospheric parameters, which in turn can influence mineral dust processes again.Combining the above topics of dust and its interaction with CPOs and radiation, the influence of the radiative effect of dust contained in CPOs is investigated as a part of this study.The modification of the atmospheric radiation budget by CPOs has been shown to be of importance by Redl et al. (2016).They investigate the influence of the radiative effect of moisture contained in CPOs on boundary layer dynamics.However, a shortcoming of their study is the missing radiative effect of mineral dust, which is often emitted by CPOs in desert regions.This is also a shortcoming in the study of Heinold et al. (2013) who simulate dust emission due to CPOs but do not include the mineral dust radiative effect.Kalenderski and Stenchikov (2016) include and investigate the mineral dust radiative effect in their convection permitting simulation of a CPO in the Red Sea region.However, they do not publish systematic findings on its feedback on the CPO structure.
Consequently, the research questions addressed are as follows: (1) Is the forecast of the dust event improved by running convection permitting simulations?(2) How does the synoptic situation relate to its exceptional character?(3) What are the meteorological drivers responsible for pick-up and long-range transport of mineral dust?(4) How does the mineral dust radiative effect influence the dust event in general and the evolution of the CPOs in particular? 2 Model description ICON is a non-hydrostatic modelling system developed jointly by the German Weather Service (DWD) and the Max Planck Institute for Meteorology (Zängl et al., 2015).It solves the full three-dimensional non-hydrostatic and compressible Navier-Stokes equations for all domains.Thereby, ICON can serve as a unified global numerical weather prediction model and climate modelling system, enabling seamless prediction from the global to local scale with a unified set of model physics.Its major advantages over previous model generations used at DWD and especially important for atmospheric tracer studies are the exact local mass conservation achieved by solving a prognostic equation for density, and the mass-consistent tracer transport achieved by transporting time-averaged mass fluxes computed from the dynamical core and diagnostic reintegration of the mass continuity equation.ICON allows for flexible local grid refinements (nests) with two-way interactions between the respective grids.Furthermore, it features a better scalability on massively parallel computer architectures.Since January 2015 ICON is used for operational weather forecasting at DWD.The ART module is an extension of ICON developed at the Institute of Meteorology and Climate Research at the Karlsruhe Institute of Technology.An overview of the module is given by Rieger et al. (2015).ART is capable of simulating a variety of aerosol species, e.g.volcanic ash, sea salt and radioactive substances.
In addition, atmospheric chemistry processes, e.g. two species of very short-lived bromocarbons, a linearised ozone chemistry and photolysis are also available in ART.For the tracer transport simulations the seamless modelling capabilities of ICON are of crucial importance because inconsistencies in tracer transport and tracer physics at the nest boundaries can be avoided, which is a major problem for other modelling systems.
The size distribution of mineral dust is represented by three modes in ART.For each mode the integral values of specific number and mass are the prognostic variables.The distribution of specific number and mass with particle size during transport is described using log-normal distributions for each mode with the diagnostic median diameter of the mass distribution and constant geometric standard deviation as parameters (Mode A, d = 1.5 µm, σ = 1.7;Mode B, d = 6.7 µm, σ = 1.6;Mode C, d = 14.2 µm, σ = 1.5).The processes which affect mineral dust number and/or mass concentrations in ART are sedimentation, dry deposition and wet deposition due to washout.The scheme used for emission of mineral dust in ART is described in Rieger (2016).It is based on an emission scheme introduced by Vogel et al. (2006).Compared to the original version three improvements were implemented, these are (1) the global availability of soil properties (size distribution, residual soil moisture), (2) accounting for the soil dispersion state, and (3) a tile approach used to account for soil type heterogeneity at coarse resolutions.
For the equations used the reader is referred to the work by Rieger (2016).The scheme parametrizes the threshold friction velocity above which dust emission can occur according to Shao and Lu (2000).They base their description on a physical balance between aerodynamic drag and lift causing upward directed forces and cohesion and gravity causing downward directed forces on particles.The effect of surface roughness and soil moisture are accounted for through parametrizations by Raupach et al. (1993) and Fécan et al. (1999) respectively, which modify the threshold friction velocity.The regional distribution of soil types used for dust emission in ART is taken from the Harmonized World Soil Database (HWSD) dataset with a resolution of 30 arc seconds (Nachtergaele and Batjes, 2012).The fraction of erodible soil is determined assuming that certain land use classes from the GlobCover2000 dataset (Arino et al., 2008) (Gleick, 2014), problems in transboundary water management (Voss et al., 2013) and a drought period (Notaro et al., 2015).Therefore, widespread, previously cultivated areas exist which are now available to erosion and known to be efficient dust sources (Solomos et al., 2016).These recent land cover changes are not reflected in the GlobCover2000 land use dataset and can result in underestimation of dust emission.

Mineral dust radiative effect
As a part of this study the on-line dust radiative effect has been implemented in ICON-ART.It is now possible to include the radiative effect of the current, local dust concentration from ART at every grid point and time step in ICON instead of the previously used dust climatology.Through its feedback on radiative fluxes the dust influences atmospheric state, thereby providing a feedback loop back to dust processes again (Tegen et al., 2006;Heinold et al., 2008;Shao et al., 2011a).The implementation was done for the standard radiation scheme utilized by ICON which is the Rapid Radiative Transfer Model (RRTM) described by Mlawer et al. (1997).In our setup, the RRTM is called every 288 seconds for the finest resolution in ICON.
The ART dust radiation routine is called at every time step at which the RRTM is called by ICON.Situated at the interface between ICON and the RRTM, ART modifies the radiative transfer parameters of the climatological dust distribution.Specifically, at every grid-point and for every level the optical depth, single scattering albedo and asymmetry parameter are calculated from the dust optical properties as a function of the dust mass concentration as detailed in Stanelle et al. (2010).The parameters returned by ART are the combined values from the local ART dust concentration plus the Tegen climatology, the latter being used only for aerosol species not simulated in ART.
In order to include the on-line dust radiative effect by modification of the radiative transfer parameters the dust optical properties have to be determined.The mineral dust optical properties are computed with the help Mie calculations using the complex refractive index of mineral dust (Bohren and Huffman, 1983;Petty, 2006;Wagner et al., 2012).This study therefore assumes sphericity as well as a spatially invariant mineral composition of the mineral dust particles.Although this assumption is generally not fulfilled for single mineral dust particles (Otto et al., 2009;Kahnert et al., 2007), it introduces negligible errors for a population of randomly oriented non-spherical particles if only albedo and flux related quantities are calculated (Mishchenko et al., 1995(Mishchenko et al., , 1997)).Due to the random orientation a collection of particles scatters light similar to a spherical particle because the individual differences disappear in the angular integration (Tegen and Lacis, 1996).Therefore Mie calculations are able to provide a good representation of scattering even for non-spherical particles.The values of the refractive index used to conduct the Mie calculations are the same ones used by Stanelle et al. (2010) for COSMO-ART and therefore not detailed further.Atmos.Chem. Phys. Discuss., doi:10.5194/acp-2017-11, 2017 Manuscript under review for journal Atmos.Chem.Phys.Published: 3 February 2017 c Author(s) 2017.CC-BY 3.0 License.
For the Mie calculations a code developed by Bond et al. (2006) was used, this in turn utilizes a subset developed by Mätzler (2002) for calculation of the Mie scattering coefficients and truncation of the series.The code was adapted to allow for processing of multiple wavelengths and averaging to the RRTM wavebands in a post-processing step.The mineral dust optical properties are calculated for three modes and 30 RRTM radiation wavebands, respectively.Results of Mie calculations for the ART mineral dust modes are shown in figure 2 and are comparable to those published by other studies (Tegen and Lacis, 1996;Helmert et al., 2007;Rémy et al., 2015).Although the authors use slightly different size distributions and refractive index properties the results are very similar to the ones presented here.
The optical properties of mineral dust are highly dependent on its particle size distribution represented through three lognormal modes in ART.From the six prognostic variables in ART, the specific dust mass and number concentration for each of the three modes, a median diameter can be diagnosed for each mode.Due to different processes such as sedimentation and dry deposition acting differently on the specific dust mass and number concentrations in ART, the diagnostic median diameter of each mode changes during transport (the standard deviation of each mode is kept constant).In a physical sense, the most important effect is large particles settling out faster due to sedimentation, which results in mostly smaller particles being transported to distant regions within each mode.Therefore, the median diameter of each mode is expected to decrease during transport.Consequently, a new polynomial parametrization of the optical properties per waveband was introduced to account for the change in median diameter during the transport processes.In a post-processing step a third order polynomial is fitted to the results of multiple Mie calculations for median diameters between 0.25 − 1.25 times the initial median diameter of the count number distribution.This is done for every mode and every RRTM waveband respectively.The polynomial fit parameters are initialized in ICON-ART and used for determination of optical properties at every grid point with the diagnosed median diameter being the independent variable.
The specifics of the mineral dust radiative effect implementation are available upon request and detailed in Gasch (2016).

Model set-up
In this study ICON-ART is run in a set-up with one global domain and four nests, with two-way feedback for the meteorological parameters enabled for all domains.Each domain presents a stand-alone model run which obtains its lateral boundary conditions from the coarser domain.For the global domain an R2B6 grid is used, this corresponds to an effective grid spacing of 40 km (Zängl et al., 2015).For every nest, the grid point distance is halved, thereby ending at R2B10 with an effective grid spacing of 2.5 km.The finest domain is circular shaped and centred at 35 • N, 40 • E with a radius of approximately 1000 km, thereby containing 507100 grid points.In order to realistically represent organized moist, deep convection a convection permitting model resolution is required.A grid spacing of 2.5 km is generally assumed to be sufficient to permit the development of convection in a non-hydrostatic model.Therefore the convection parametrization, including the parametrization for shallow convection, is switched off for the finest grid.
On the global domain the model consists of 90 levels extending up to the mesosphere, with the lowest level being at 20 m and the highest level at 75 km.For the nests the simulated atmosphere extends into the stratosphere up to 22.5 km containing 60 vertical levels.
For the cloud micro-physical processes a the two-moment cloud scheme is used (Seifert and Beheng, 2006), as this was found to lead to more realistic features of the meso-scale organized convection.The two-moment scheme utilizes a parametrization developed by Seifert and Beheng (2001) which predicts number and mass concentrations for six different hydro-meteor species.
These are cloud droplets, rain drops, cloud ice, snow, graupel and hail.
ICON-ART is initialized with analysis fields from the Integrated Forecasting System (IFS) of the European Centre for Medium-Range Weather Forecasts (ECMWF).A limitation with initializing from the IFS analysis datasets is that the IFS has a horizontal grid spacing of approximately 13 km and is therefore non convection-permitting.Thus, when re-initializing ICON-ART with the current meteorological fields any previously existing organized convection is terminated.The IFS initialization data for soil moisture was modified in a region along the Syrian-Iraqi border which showed unrealistically high soil moisture values and spatial patterns without preceding rain or changes in soil properties.Therefore, in a region from 37.5 UTC 06 September onwards, a multitude of partially overlapping events take place which need to be simulated correctly in order to obtain a realistic dust distribution in the EM.Therefore, the time chosen for initialization is of crucial importance due to the aforementioned termination of organized convection.Various options were examined, the times chosen for initialization are 18 UTC 05 September and 12 UTC 06 September.
For investigation of the dust radiative effect two simulations are performed.The first is the simulation including the on-line radiation interaction with mineral dust from ART which is called 'ARI' (aerosol radiation interaction) in the following.The second is a simulation in which the mineral dust concentrations are multiplied by zero in the ART routine calculating the radiative effect of dust.This simulation is called 'CTRL' (control) in the following, it contains no mineral dust influence on radiation at all.In this study only the mineral dust radiation interaction is simulated on-line by ART, for all other aerosols the default climatologies are used in all runs.

Simulation results and validation
The ICON-ART simulation results show that the dust event consists of multiple stages and is created by the interaction of different meteorological systems.A comparison of model results to available satellite observations highlights the ICON-ART simulation quality.

Synoptic situation
For the event simulated, the synoptic conditions in the Middle East are distinctively different from the normal summer situation as the ICON-ART model results show.An overview of the modelled synoptic situation is provided in figure 3 at 18 UTC 06 September, as at this point in time all relevant synoptic features are detectable.
The usually dominant Persian trough (Bitan and Sa'Aroni, 1992) does not exist.Instead a low-level synoptic pattern termed Red Sea Trough (RST) with an eastern axis is apparent below 800 hPa which extends northward from the Red Sea towards central Syria under easterly flow conditions.The direction of the trough axis and its position over the Red Sea make it plausible to attribute it to the RST category rather than to the Persian trough category.Furthermore, a characteristic strong southerly flow exists in the Mesopotamia region, which is crucial for the sustained lifetime of the meso-scale organized convection as it advects hot and moist air from the Persian Gulf.The RST is accompanied by a mid-tropospheric trough between 700−400 hPa extending far southward into the EM under westerly flow conditions.The trough axis runs approximately along the Dead Sea Rift Valley, curving eastwards towards Egypt in the southern part.East of this trough (downstream) high wind speeds up to 20 m s −1 exist in a streak running from south-west towards north-east.The streak is situated exactly upstream of the mesoscale convective system (MCS) development region.As suggested in previous studies (Krichak et al., 2012;Vries et al., 2013),  In combination with the orographic lifting by the Zagros mountain range, the position of the slowly eastward moving trough enables the development of meso-scale organized convection and determines its position throughout the event.The combination of a low-level RST extending northward and upper tropospheric troughs extending southward into the EM has been termed active RST in previous studies due to its high potential for severe weather (Krichak et al., 2012;Vries et al., 2013).The timing of the active RST synoptic situation is exceptional, as the RST usually starts to occur only by late September or early October (Alpert et al., 2004;Tsvieli and Zangvil, 2005), although events have been documented in August and September (Osetinsky, 2006).The active RST has been linked to severe weather phenomena in the EM and the Arabian peninsula in connection with atmospheric rivers transporting large quantities of precipitable water from eastern Africa, although the specific moisture sources are still debated (Vries et al., 2013, and references therein).The clustering of convectively active days with meso-scale convective organization is often observed during active RST situations (Krichak et al., 2012;Vries et al., 2013).This clustering is also in agreement with a study by Miller et al. (2008), who investigate haboob characteristics in the Arabian peninsula.
As is discussed in Sec.3.2, the active RST enables the interaction of multiple dust emitting meteorological systems over the course of three days, which explains the extraordinariness of the event with respect to magnitude and spatial extent.The unusual transport direction of the dust plume from Syria and Iraq into the EM from the east is caused by the downstream flank of the RST.To our knowledge, the active RST synoptic situation has previously not been linked to severe dust events in the EM and the exceptional character of this event is emphasized by a comparison with climatological studies (Singer et al., 2003;Alpert et al., 2004;Dayan et al., 2008).

Course of events
In the following, a detailed analysis of the development stages and responsible atmospheric drivers which lead to the severe dust event is provided.The simulated convection and its interaction with dust emission is investigated in depth.Unfortunately, the region where the MCS and the first stages of the CPOs occur are located in the Syria-Iraq border region, which is not covered by a meteorological observation network.Therefore, no surface observations are available and the event can only be analysed using satellite data, which nevertheless yields interesting results.Figure 4 provides a schematic depiction of the event stages as well as their horizontal extent and course which are referred to in the following.

First cold-pool outflow, heat low and Eastern Mediterranean sea-breeze
During the night from 05 to 06 September 2015 a convective system exists over the Turkey-Syria border region.It is fuelled by the inflow along the eastern side of the RST.The system moves towards the north-east along with the mean flow direction above 500 hPa.Due to the favourable position of the convective system in front of the mid-tropospheric through axis, it Dust pick-up in CPO2 and CPO3 is limited during the night but increases as soon as boundary layer mixing increases due to solar insolation.At 10 UTC 07 September all satellite platforms are available and an in-depth analysis is conducted.

Meso-scale convective system and cold-pool outflows
An in-depth analysis of model results and comparison to satellite observations is conducted at 10 UTC 07 September (fig.5).
The  The structure of the MCS determines the shape of the outflow, therefore differences in outflow structure can be identified as well.Again, the CPO3 is more sharply confined between 34 For the eastern part of CPO3 the MODIS AOD measurements seem doubtful when comparing to the MODIS visible satellite image (although it is reduced by a value of 0.3 to account for other background aerosols measured by MODIS, the discrepancy is much larger than this).
The aged CPO2 spreading towards the north is detected by all satellite instruments, it is modelled with an offset towards   as a thin line in 2m-DPT, a gradient in DOD and a change in wind velocity along the front (as can be expected from a frontal structure).ICON-ART is able to simulate the re-intensifying sea-breezes, thereby enabling the above analysis, however no dust pick-up is connected with the frontal structure.This is due to the less emissive soil type in the region passed by the front.In reality, however, dust was picked up by this frontal structure, this is detectable in all three satellite measurements (fig.6).As a result, CALIOP reports high values of attenuated backscatter in this region whereas ICON-ART simulates a minimum due to 5 the clean air characteristics of the sea breeze.
The main CPO3 dust plume is distinct in both figures with ICON-ART simulating an elevated nose of the CPO3.The above discussed differences in CPO3 structure are identifiable again.ICON-ART simulates a wider and more shallow outflow whereas in reality it was more confined.In addition, the observations show an approximately one kilometre higher main dust plume compared to ICON-ART.Altitudes below 2 km are marked as no signal regions in the CALIPSO feature mask due to

Long-range transport to the Eastern Mediterranean
From 10 UTC onwards on 07 September, the MCS starts to dissipate and is in the course advected east along with the mean flow conditions above 500 hPa.The CPO3 airmass created during the night spreads south-west during daytime on 07 September.
At all times it shows the characteristic features associated with CPOs such as an increase in surface wind speed, higher dew point temperatures and an arc cloud forming above the leading edge during the afternoon hours.The high surface wind speeds and turbulent mixing inside the CPO3s result in enormous dust emissions during daytime, consequently the dust is transported within the full boundary layer height up to 5 km.
With nightfall on 07 September the CPO2 and CPO3 merge.As the merged CPO is still located in the western, downstream flank of the RST the air mass and dust contained within it are advected fast towards south-west over the course of the night.
The Dead Sea Rift Valley is passed by the dust mass after midnight on 08 September, with the dust plume interacting with the complex orography.A comparison of simulated dust optical depth with satellite observations for 11 UTC 08 September shows that ICON-ART values are one order of magnitude higher than other simulation results in the northern part of the EM (see Sec. 1 and fig.A4).However, dust transport across the Dead Sea Rift Valley into the southern EM is not simulated with the correct magnitude by ICON-ART despite the overall dust plume structure being captured well.Therefore, the transport into this region is analysed further in section 3.4.1.Overall, a significant dust forecast improvement is achieved through convection permitting simulations with ICON-ART at 47 hour forecast time without data assimilation.During daytime on 08 September the dust plume is mostly stationary in the EM and influenced by the local circulation systems.In visible satellite pictures, the dust can be seen to remain in the EM at high concentrations over the course of the next four days.This period is not investigated as a part of this study as the scope of this work is the analysis of the generating mechanisms.
Due to the problems in simulating dust transport towards the southern EM with the correct magnitude, the next section investigates the timing and structure of the CPO arrival in Israel in reality and simulation.This reveals the interaction of the dust plume with the orography in the region in the form of super-critical flow conditions and connected, subsequent hydraulic jumps.

Environmental station data
In this section the simulated dust concentrations are compared to measurements from three stations in Israel.The selected stations are Afula, Jerusalem (Bar Ilan) and Ashdod (Nir Galim) as they all show individual dust concentration characteristics during the event (see fig. 9 c) for a map of the station locations).
The modelled dust plume reaches Afula first, this is also where the highest values of DOD up to 1.5 are achieved because of the previously discussed transport through the Golan Heights (first row in fig.8).A few hours later, Jerusalem shows an increase and its peak optical depth.DOD increases even later in Ashdod due to its location in the south-west, but the DOD values reach higher levels than in Jerusalem due to the lower altitude of the station.
The large difference in absolute values of modelled and measured PM10 concentrations (second and third row in fig.8) is caused by the inadequately modelled interaction of the dust plume with the complex orography of the Dead Sea Rift Valley as the next section shows.
Despite the large difference in absolute values, the dust transport processes can be studied by comparing the course of the measured and modelled PM10 concentrations.Although the DOD in Afula reaches highest values during the night already, peak concentrations at 57 m surface elevation are only measured after noon on 08 September.This is due to arrival of the dust plume in 1 km height.Consequently, the onset of turbulent mixing is needed to transport the high dust concentrations towards the surface.The discussed processes are modelled sufficiently by ICON-ART, as it shows a very similar shape of the PM10 concentrations curve compared to the measurements with a peak at 12 UTC.
Highest overall PM10 concentrations are measured by the Jerusalem station.It shows the earliest and most pronounced peak of all stations at 08 UTC.This is also the case in ICON-ART, although with a shorter peak duration.The earlier and higher peak concentrations in Jerusalem are to due to the elevated location of the measurement station at 770 m which is almost inside the dust plume.For this reason, PM10 concentrations are also more correlated with the shape of the DOD curve at this station.
A secondary peak after sunset is visible in the measurements and model results.This peak can be explained by the formation of a stable nocturnal boundary layer, in which the dust from higher altitudes settles to the ground.

Hydraulic jump upstream the Dead Sea Rift Valley
In order to understand the failure of the model to forecast dust in Israel with the correct magnitude the transport processes upstream have to be investigated in more detail.
In figure 10 a cross-section of ICON-ART results along a cross-section through the Golan heights is displayed for 03 UTC 08 September.The location of the cross-section and according meteorological situation is illustrated in figure 9.The arrival of 5 the dust plume is simulated above a height of 1 km in Israel.This is in good agreement with measurements of the dust plume height in Israel which also report an arrival of the dust plume in 1 km height (Alpert et al., 2016).
Remarkable is a wave structure in the extinction coefficient on the lee side of the Golan Heights (fig.10).In the region of the wave structure, after passing the Golan Heights crest, the vertical depth of the dust plume decreases significantly to a few hundred meters.The plume expands and returns to its original height again in an abrupt expansion further to the west which   Hydraulic jumps are connected to flows going from a sub-critical flow stage with Froude numbers F r smaller one, to a super-critical flow stage, with Froude numbers greater one.According to Drobinski et al. (2001) the Froude number can be calculated as: Thereby, U denotes the wind speed and Θ v the virtual potential temperature in the atmospheric boundary layer, chosen at a representative level and ∆Θ v the temperature inversion at the boundary layer top.Further, g is the gravitational acceleration and h the atmospheric boundary layer height.
Hydraulic jumps have been documented penetrating the Dead Sea Valley for flow conditions from the west (Metzger, 2016).
The flow across the Golan Heights shows characteristic features of a super-critical flow with a subsequent hydraulic jump (see figure 10).These are: 1.A continuous decrease in mixing layer height detectable in the virtual potential temperature and extinction coefficient fields leading to a compressions of the sub-critical flow from east to west.An approximate calculation of the Froude number upstream and downstream the Golan Heights crest height with average flow conditions as simulated by ICON-ART confirms the conversion of the flow state: Hydraulic jumps have been investigated and suggested as dust generating mechanisms (Cuesta et al., 2009).Their existence and interaction with the orography under strong easterly flow conditions has been proven for the Sahara (Drobinski et al., 2009).As the gust front of CPO3 reaches the Dead Sea Rift Valley almost along its full length, the flow reaches supercritical stage in the vicinity of many prominent orographic features (fig.9 b)).Consequently, the widespread existence of hydraulic jumps in connection with the upstream super-critical flow is assumed to be responsible for dust emissions on the eastern side of the Dead Sea Rift Valley.Dust emission in the region is assisted by the timing of the event, as in September the soil in the region is most erodible due to the preceding hot and dry summer.
The existence of hydraulic jumps is modelled by ICON-ART and detectable along many cross-sections and at different times in the region.Some of them occur on the lee-side of much less pronounced orographic features such as the mountain structure visible at 33.5 • N, 37.0 • E in figure 10.Although ICON-ART simulates the flow phenomena, the location and magnitude of the hydraulic jumps need to be validated in order to evaluate the simulation quality.

Meteorological station measurements
The validation is done with the help of three meteorological stations approximately along the model cross-section through the Golan Heights.The locations and names of the stations used are marked in figure 9 c).Unfortunately, no measurements are available in Jordan, where the existence of the hydraulic jumps is detectable in many places in ICON-ART and where the more erodible state of the soil is assumed to be responsible for high dust emissions.underestimated by an order of magnitude.Nevertheless, ICON-ART provides us with a detailed understanding of previously unknown processes contributing to the historic dust event which makes these findings worthwhile to report.

Mineral dust radiative effect
The validation of the mineral dust distribution and transport characteristics with satellite and station measurements show overall good agreement between ICON-ART and the observations, especially during the early stages of the event.The simulation therefore allows for investigation of the mineral dust radiative effect.Using ICON-ART, the radiative effect of mineral dust can be studied in detail through the differences between the simulation including mineral dust radiation interaction (ARI) and without it (CTRL).An analysis of the mineral dust radiative effect on surface conditions is conducted at two characteristic points in time during the dust storm development.
The first analysis is performed at 03 UTC 07 September as it shows interesting night-time features due to mineral dust (fig. 12).The point in time chosen is just before sunrise in the region.The mineral dust affected areas show the expected behaviour in terms of its influence on radiative transfer.Net global radiation at night is increased by values between 5 − 50 W m −2 , with higher values in regions with higher aerosol optical thickness (fig.12 d)).The increase is caused by an increase in down-welling long-wave atmospheric radiation (not shown).Consequently, the amount of energy radiated into space is reduced, which causes a night-time warming of the surface.The reductions in net global radiation towards the east are caused by the daytime effect of mineral dust, as the sun is rising in this region already.The 2m-temperature field outside the CPO3 and MCS region (which themselves are discussed further below) shows the expected behaviour for mineral dust at night with an increase in surface temperature for ARI (fig.12 c)).Also visible are smaller scale irregularities showing local temperature differences located outside the region affected by mineral dust.These are introduced due to the slightly different location and timing of convection in both runs.The magnitude of the temperature response to mineral dust radiative forcing is large when compared to other studies with values up to 5 K (Tegen et al., 2006;Highwood and Ryder, 2014;Rémy et al., 2015).A possible explanation are the high values used for the imaginary part of the mineral dust refractive index in the near infra-red spectral region, making the The reductions in surface temperature have a stabilizing effect on the boundary layer (Heinold et al., 2008).Is the forecast of the dust event improved by running convection permitting simulations?
The convection permitting simulation of the dust event with ICON-ART improves the forecast quality decisively.The driving meteorological systems and resulting dust emissions are captured in their horizontal, vertical and temporal structure as is shown

Figure 1 .
Figure 1.MODIS VIS satellite images of the EM region, from left to right for 06, 07 and 08 September 2015 (NASA Worldview, 2016).
contribute to mineral dust emission whereas others do not, with snow Atmos.Chem.Phys.Discuss., doi:10.5194/acp-2017-11,2017 Manuscript under review for journal Atmos.Chem.Phys.Published: 3 February 2017 c Author(s) 2017.CC-BY 3.0 License.generally prohibiting emissions.The land use classes which can contribute to emission are regions with sparse vegetation, bare areas, closed to open grass-and shrub-lands, furthermore mosaic forest/grassland and shrub-land.In order to retain the high spatial resolution of the dataset for ART a tile approach is used for calculating dust emissions.The tile approach calculates the overall emission in every grid box as a weighted average of the emissions from different soil types based on their fractional coverage of the grid box.Determination of land use classes is difficult in the Levantine region.It has undergone drastic change in recent years due to on-going conflict

Figure 2 .
Figure 2. Results from Mie calculations as used by ICON-ART.Points are normalized with respect to value at 550nm, horizontal lines show absolute values.Shown are from top to bottom: Extinction coefficient (EXT) in m 2 g −1 , single scattering albedo (SSA) and asymmetry parameter (ASY) for all modes.The borders of the RRTM radiation scheme wavebands are adumbrated as grey lines in the background.The filled grey band represents the waveband present both in the longwave and shortwave part of the RRTM.
• N − 41.5 • N and 32.5 • E − 35 • E the soil moisture index in the four layers provided by the IFS is set to the average value of the region between 36.5 • N − 38 • N and 32 • E − 34 • E which represents the soil moisture conditions in the region more realistically.The region modified is an important dust source region and emission fluxes for mineral dust increased due to the reduction of the soil moisture content.The ICON-ART mineral dust dust concentrations are passed on to the next run whenever a reinitialization of the meteorological fields from an IFS analysis is performed.No assimilation of mineral dust concentrations from observations takes place.Hence, the runs are performed as free runs for the mineral dust concentrations with a frequent update of the meteorological Atmos.Chem.Phys.Discuss., doi:10.5194/acp-2017-11,2017 Manuscript under review for journal Atmos.Chem.Phys.Published: 3 February 2017 c Author(s) 2017.CC-BY 3.0 License.background conditions from IFS analysis in order to benefit from the data assimilation performed therein.A two week spin-up simulation is performed starting on 23 August 2015 in order to achieve a realistic background concentration for mineral dust on the global domain.Another reinitialization from an IFS analysis is performed at 00 UTC 04 September to obtain a realistic background concentration of mineral dust in the finer domains for the simulations starting at 18 UTC 05 September.From 00 Atmos.Chem.Phys.Discuss., doi:10.5194/acp-2017-11,2017   Manuscript under review for journal Atmos.Chem.Phys.Published: 3 February 2017 c Author(s) 2017.CC-BY 3.0 License.

Figure 3 .
Figure 3. Synoptic situation on 06 September at 18 UTC as simulated by ICON-ART for the global domain.Shown are from top to bottom the 300 hPa, 600 hPa and 900 hPa level.Colour coded on the left side is the relative topography as the geopotential height of the respective layer minus the geopotential height of the 1000 hPa layer.Black lines denote the height of the respective pressure level in geopotential metres at 5 gpm intervals.White lines show surface pressure in hPa.On the right side wind speed is colour coded and wind velocity is shown as vectors.

Figure 4 .
Figure 4. Schematic depiction the main dust event stages and atmospheric features.Colour-coded is the topography as used by the finest domain.Coloured lines represent average geopotential height of pressure levels.Labels refer to the description used in the text.Arrow size corresponds to event magnitude and arrowhead size to event speed.
simulated MCS has passed its most intensive development stage approximately four hours earlier.The main MCS features are still visible, although it is in the stage of dissolution due to the shift of the upper atmospheric trough towards east (fig.5 c)).The convection is organized along the orographic features of the Zagros mountain range, exhibiting a sharp line shaped rainfall distribution.Possibly, the convective structure is also shaped by a weak cyclogenesis taking place further east due to the favourable upper-tropospheric conditions.The cirrus cloud anvil extends far to the north-east as it is transported away by the upper level flow.Downstream the line-shaped rainfall distribution near surface wind speeds increase strongly as the CPO3 reaches the surface (fig.5 b)).Wind speeds above 12 m s −1 are modelled inside the CPO3 region.The southern edge is distinct and counters the inflow from the Persian Gulf region.The edge is also clearly recognizable in the 2m-dew point temperature (2m-DPT) field which is a good indicator for CPOs due to the change in air mass characteristic (Knippertz et al., 2007).The difference in 2m-DPT between the CPO3 and the surrounding air masses is approximately 10 • C (fig. 5 d)).Dust has been picked up in large quantities over the previous hours with increasing boundary layer turbulence and dust optical depth (DOD) values above 2 are modelled in the CPO3 region (fig.5 a)).The maximum DOD value is 4.15, it is reached in an area close to the leading edge of CPO3 which shows the highest values of DOD.Towards north, the DOD exhibits a sharp gradient (northern boundary of CPO3 dust plume in fig.5).At closer inspection of the wind and dew point temperature fields it becomes apparent that the decrease in DOD is not linked to the extent of the CPO3 in north-western Syria (fig.5 b) and d)).This is due to the orography and the soil-type distribution in the region, a change in soil type towards less-emitting light clay and the increase in elevation are responsible for the reduced DOD towards north.The northern edge of the aged CPO2 is marked by a line of convective clouds over the Taurus mountain range, here a second but less distinct gradient in DOD is visible.In the south-western part of Syria, entering Jordan, the remains of CPO2 are also still Atmos.Chem.Phys.Discuss., doi:10.5194/acp-2017-11,2017   Manuscript under review for journal Atmos.Chem.Phys.Published: 3 February 2017 c Author(s) 2017.CC-BY 3.0 License.

Figure 5 .
Figure 5. ICON-ART model results at 10 UTC 07 September.Displayed are: a) Dust optical depth at 550 nm overlain with column integrated hydro-meteor content TQ (cloud water, cloud ice and graupel).b) Wind speed at 385 m model height and wind velocity as vectors.c) Fractional cloud cover and surface rain rate colour coded.Blue lines show geopotential height of 900 hPa level at 10 gpm intervals, green lines show geopotential height of 500 hPa level at 20 gpm intervals.d) 2m-dew point temperature.From west to east solid black lines mark leading edges of CPO1, CPO2 and southern CPO3 boundary.Gray line marks northern dust plume boundary inside CPO3.Dashed line marks merged EM Sea-breeze front from 05 and 06 September.Dotted line marks frontal region of inflow.
the north-east by ICON-ART (northernmost solid black line in fig.6).Noticeable are arc clouds forming above the Taurus mountains in simulation and reality.The south-western boundary of the aged CPO2 is modelled correctly in the Golan Heights region.However, towards the west and south the CPO2 has advanced further in reality than in the model, as all measurements show a southern edge of the CPO2 which is just about to cross the border into Saudi-Arabia.Aerosol optical depths retrieved by MODIS are in the range of 1.5 whereas ICON-ART simulates values in the range of 1.In this case, the MODIS measurements seem more realistic when compared to the visible satellite picture.A possible reason for the underestimation of DOD magnitude in the model are reduced dust emissions due to an unrealistic land use characterisation.Due to recent changes in the region (see Sec. 2) an update of the land-use classes can be expected to yield significant improvements of modelled DOD magnitude.The vertical structure of the dust plume can be investigated at this point in time with the help of a CALIPSO overpass which occurred at 10.35 UTC.Results along the flight track are shown from south to north in figure 7. The southern inflow region towards the MCS shows high values of the extinction coefficient in ICON-ART and attenuated backscatter in CALIOP measurements.The height of the dust plume is between 5 − 6 km, with decreasing height towards the MCS.Just south of the main CPO3 dust plume, in a region from 32 − 33 • N both show a decrease in dust concentrations at upper levels.However, a difference exists in the near surface values up to 2 km, where CALIOP reports high values of total attenuated backscatter but ICON-ART shows a minimum in extinction coefficient.The reason for this difference becomes clear when looking at the CALIPSO ground track again in figure 6.The satellite passes the region where the merged one and two day old sea-breezes from the Mediterranean penetrate the RST inflow in a cyclonic rotational movement.The frontal structure is visible in figure 5 Atmos.Chem.Phys.Discuss., doi:10.5194/acp-2017-11,2017 Manuscript under review for journal Atmos.Chem.Phys.Published: 3 February 2017 c Author(s) 2017.CC-BY 3.0 License.

Figure 6 .
Figure 6.ICON-ART model results and satellite observations for 10 UTC on 07 September.Axis for dust optical depths, column integrated hydro-meteor content and event marks as in figure 5. Displayed are: a) ICON-ART dust optical depth, column integrated hydro-meteor content and wind velocity.b) SEVIRI RGB dust product (Kerkmann et al., 2015).c) Aqua MODIS VIS satellite image, overpass at 10.35 UTC (NASA Worldview, 2016).d) Aqua MODIS aerosol optical depth retrieval using DB2 algorithm (AOD-0.3)(Levy and Hsu, 2015), overpass at 10.35 UTC.Chain dotted black line marks CALIPSO ground track at 10.35 UTC.

Figure 7 .
Figure 7. Vertical cross-section of ICON-ART model results for 10 UTC on 07 September and CALIPSO satellite observations from south to north along ground track in figure 6. Left side: ICON-ART mineral dust extinction coefficient.Right side: Total attenuated backscatter as measured by CALIOP.No quantitative comparison is possible as two different measures are displayed.Southern dotted black line marks inflow region.Dashed black line marks aged EM Sea-breeze penetrations region.Southern solid black line marks CPO3.Northern solid black line marks aged and lifted CPO2.

10
the attenuation of the lidar signal (not shown).On the northern side of the plume beyond 36 • N a minimum captured by both is visible, once more illustrating the well simulated northern boundary of the CPO3 in this region.Further towards the north the 16 Atmos.Chem.Phys.Discuss., doi:10.5194/acp-2017-11,2017 Manuscript under review for journal Atmos.Chem.Phys.Published: 3 February 2017 c Author(s) 2017.CC-BY 3.0 License.overarching dust plume structure consisting of dust picked up by the aged CPO2 is represented very similarly in simulation and measurement.Summarizing, a northward deviation of the flow structure in ICON-ART can be identified both for CPO2 and CPO3 although the overall intensity and characteristics are simulated well.The combination of the northern deviation towards less dust emitting soil types and the wider, less well defined main outflow region reduce the amount of DOD over western Syria due to a less intense channelling of the CPO3.Furthermore, the re-intensification of the aged CPO2 is modelled by ICON-ART, however, it was observed with greater magnitude in reality.The reduced amount of dust in the atmosphere in ICON-ART leads to a reduction in long range transport of dust towards the EM over night.

Figure 8 .
Figure 8. Observations and model results for three stations in Israel for 07 and 08 September.Points denote observations and solid lines model results.Please note the difference in y-axis scaling by a factor of ten between the second and third sub-plot.Shown are from top to bottom: 1. Modelled dust optical depth 2. Measured PM10 concentrations.3. Modelled PM10 concentrations.4. Global radiation (GRad), the values shown are for Afula, Bet Dagan and Jerusalem (Givat Ram) due to data availability.

Figure 9 .
Figure 9. ICON-ART model results at 03 UTC 08 September, same as figure 5 but for the Dead Sea Rift Valley area.The bottom left plot shows the orography in the region colour-coded.The dashed black line marks the region where super-critical flow conditions and the subsequent hydraulic jump occur.The solid black line in c) shows location of ICON-ART vertical transect displayed in figure 10.In addition, location and names of stations used for comparison of results are displayed.

Figure 10 .
Figure 10.ICON-ART model results along a vertical cross-section from 32 • N, 34 • E to 34 • N, 38 • E at 03 UTC 08 September.Displayed on the left side is the colour-coded extinction coefficient overlain with virtual potential temperature contours.On the right side wind-speed is colour coded and overlain with turbulent kinetic energy contours.is accompanied by a sharp decrease in flow speed.The reason for this behaviour is the existence of a flow phenomena termed 'hydraulic jump'.
2. A sharp decrease in the vertical flow depth, connected to a rise in flow speed, after passing the orographic Golan Heights crest.This denotes the transformation of the flow state from sub-to super-critical.21 Atmos.Chem.Phys.Discuss., doi:10.5194/acp-2017-11,2017 Manuscript under review for journal Atmos.Chem.Phys.Published: 3 February 2017 c Author(s) 2017.CC-BY 3.0 License.3. A subsequent sudden increase in flow depth when the flow state reverts to the sub-critical stage again inside the Dead Sea Rift Valley.This is accompanied by a sharp decrease in flow speed and connected to an increase in atmospheric turbulence.

Figure 11 .
Figure 11.Observations and model results for three stations in the Golan Heights for 07 and 08 September.Points denote observations and solid lines/crosses model results.Shown are from top to bottom: 1. Modelled dust optical depth 2. Wind speed (WS).3. Wind direction (WD).4. 2m-surface temperature.5. Global radiation (GRad).
Atmos.Chem.Phys.Discuss., doi:10.5194/acp-2017-11,2017   Manuscript under review for journal Atmos.Chem.Phys.Published: 3 February 2017 c Author(s) 2017.CC-BY 3.0 License.scatters radiation compared to absorbing it, thereby converting direct radiation to diffuse radiation.As the energy reaches the surface nevertheless, reductions in surface temperature due to mineral dust are smaller than those found by to other studies.A detailed analysis for the 10 UTC situation supports this hypothesis.Decreases in direct shortwave radiation are on the order of 600 W m −2 inside the CPO3 region.However, these are countered by increases in diffuse shortwave above 200 W m −2 , giving a net reduction in short-wave radiation of approximately 400 W m −2 (not shown).Furthermore, the high amount of water vapour contained in the CPOs possibly absorbs and scatters large quantities of solar radiation without dust already, thereby reducing the effect due to mineral dust.An observation which supports this hypothesis is that the reduction in 2m-temperature does not scale linearly with the difference in DOD between inflow and outflow region of the MCS.
As a result of the increased boundary layer stable stratification dust emissions decrease almost homogeneously throughout the dust affected region (fig.13 b)).In addition, this results in a reduced maximum DOD of 4.15 in ARI compared to 5.61 in CTRL for the whole domain.During the late afternoon with maximum boundary layer development this difference increases.At 15 UTC ARI and CTRL show maximum DODs of 6.11 compared to 9.26, respectively.Another marked feature visible in figure 13 is the previously discussed difference in DOD at the CPO leading edges (fig.13 a)).As the feature is visible over more than 20 hours and in connection with all CPOs this confirms the observation of faster spreading CPOs in simulations including the mineral dust radiative feedback as proposed earlier.The streaks detectable inside the CPO3, where ARI shows higher values of DOD, can be attributed to shallow boundary layer convection which develops earlier and more intensely in the CPO3 region in CTRL due to the lacking vertical stabilization (not shown).Consequently, due to the redistribution of dust by convection the DOD field is more inhomogeneous in the CTRL run (not shown).4ConclusionsTo our knowledge, this study presents the first sucessful simulation of the September 2015 severe dust event at convection permitting resolution.An active Red Sea Trough situation and the related MCS and CPOs are shown to be responsible for the severe dust event in the Eastern Mediterranean.In addition, the interaction with an intense heat low, the land inward penetrating Eastern Mediterranean sea-breeze and the widespread occurrence of super-critical flow conditions and subsequent hydraulic jumps are suggested as important drivers for dust emission.The mineral dust radiation interaction has been implemented as a new module in ICON-ART.Based on Mie scattering calculations, which are conducted off-line to calculate the mineral dust optical properties, the radiative transfer parameters used by ICON are calculated on-line in ART to account for the mineral dust radiative effect.A new, size dependent parametrization of the mineral dust optical properties is proposed.Furthermore, to our knowledge this is the first study to investigate the mineral dust radiative effect on CPO structure.

Figure A2 .
Figure A2.ICON-ART model results and EUMETSAT satellite observations at 00 UTC 07 September.From west to east solid black mark leading edges of CPO1, CPO2 and CPO3.Dashed line marks EM Sea-breeze front from 06 September.Dotted line marks frontal region of inflow.Chain dotted black line marks CALIPSO ground track at 23.35 UTC 06 September.Left side: ICON-ART dust optical depth, column integrated hydro-meteor content and wind velocity.Right side: SEVIRI RGB dust product.

Figure A3 .
Figure A3.Vertical cross-section of ICON-ART model results at 00 UTC 07 September and CALIPSO satellite observations from south to north along ground track in figure A2.Left side: ICON-ART mineral dust extinction coefficient.Right side: Total attenuated backscatter as measured by CALIOP.No quantitative comparison is possible as two different measures are displayed.Southern solid black line marks CPO1 and HL region.Dashed line marks penetrating EM Sea-breeze.Northern solid black line marks CPO2 region.

Figure A4 .
Figure A4.ICON-ART model results and satellite observations at 11 UTC 08 September.Same as figure A1 but for Aqua MODIS instrument, overpass for the left part of the picture at 11.18 UTC.Please note that in this case the MODIS Dark Target and Deep Blue combined AOD product is displayed in order to obtain values over sea.