The AErosol RObotic NETwork (AERONET) provides valuable measurements of atmospheric aerosol optical properties. AERONET stations comprise an automatic sun photometer. It has its own algorithm to evaluate data quality and instrument functionality. AERONET measures AOD for different wavelengths, from near IR (1064 nm) to near UV (340 nm). AERONET sunphotometer provides information about aerosol size distribution, aerosol radiative forcing, and aerosol shape (spherical/non-spherical). Detailed description of the instrument and its function can be found in Holben et al. (1998).

The measurements pass multilevel quality assurance (QA): level 1 without cloud screening, level 1.5 has cloud screened but may be without final calibration. Level 2 has cloud screened and quality assured calibrations. The AOD accuracy of calibrated AERONET station is wavelength dependent and varies from root mean square error of ±0.012 (UV band) to root mean square error of ±0.006 (IR band) at overhead sun (airmass =1) (Schmid et al., 1999).

The AERONET stations are widely used over the globe. However there are only few AERONET stations in the Arabian Peninsula with relatively long measurement records. Among them is the Solar-Village station in Saudi Arabia, the Kuwait University station in Kuwait and the Mezaira site in the United Arab Emirates (UAE). For detailed information about the site and their locations see Table 1 and Fig. 1.

Each station has its unique features; the Kuwait station is downwind of major dust sources in Southern Iraq. Solar-Village is located in the center of Saudi Arabia and it is at a relatively high altitude (764 ma.s.l.) compared to other sites, which could reflect the transportation of dust to high altitudes. Mezaira in United Arab Emirate (UAE) is inland site and could be considered a receptor site for dust coming from southern Arabian Peninsula.

Most satellite instruments looks down to the earth (0 degrees Nadir angle), or toward the edge of the planet and receive the reflected sunlight. The Earth's surface, clouds and aerosols reflect sunlight in different direction, which requires an instrument that accounts for such different reflected angle. The Multiangle Imaging Spectroradiometer (MISR) is a unique instrument flown in the space since late 1998. It has nine cameras corresponding to nine view angles, the middle one pointing toward the nadir, four of them in a forward direction with (26.1, 45.6, 60.0 and 70.0∘, respectively) relative to the nadir camera and the other four in the rearward direction with (-26.1, -45.6, -60.0 and -70.0∘ respectively). Each direction measures four individual wavelength (443, 555, 670 and 865 nm). Therefore, MISR has 36 channels (Diner et al., 1998). MISR has been used in many fields including studies on, clouds (Marchand et al., 2010), aerosols (Martonchik et al., 2004; Abdou et al., 2005; Marey et al., 2011) and Earth surface (Pinty et al., 2011). For our purpose, we will focus on the aerosol facilities of MISR. The MISR global aerosol retrieval is used to obtain AOD values to characterize the types of aerosol based on their physical and optical properties, and aerosol particle shape (spherical or non-spherical). The aerosol retrieval algorithm strategy has many steps. First, it utilizes a lookup table that contains a suite of a natural aerosol types and calculated aerosol optical properties by a radiative transfer model. Second, the retrieval of the aerosol over dark surfaces like the ocean depends on red and near infrared (IR) channels (Martonchik et al.,1998). The most difficult retrieval is above bright surface like deserts. This latter retrieval needs special treatment and is therefore associated with large uncertainties (Diner, 1998; Martonchik, 1998; Abdou et al., 2005).

In this work we concentrate only on the MISR AOD dataset and its comparison with the model output. The MISR data used in our study spans 6 years from 2006 to 2012. The data resolution is 0.5∘×0.5∘. Data has been retrieved from the Giovanni website (http://Giovanni.gsfc.nasa.gov/giovanni).

The Moderate Resolution Imaging Spectroradiometer (MODIS) instrument is aboard the NASA EOS (Earth Observing System) Terra and Aqua satellites and began transmitting date in 2000. The instruments have high spatial resolution (10 km resolution) and almost global coverage. Like MISR, MODIS aerosol retrievals are based on a lookup table procedure in which the satellite measured radiances are matched to pre-calculated values in the lookup table. The values of the aerosol properties used to create the calculated radiances are retrieved (Remer et al., 2005; Abdou et al., 2005). The MODIS data set spans from January 2008 to December 2011. Data has been retrieved from the Giovanni website (http://Giovanni.gsfc.nasa.gov/giovanni).

The Ozone Monitoring Instrument (OMI) has been orbiting the Earth on one of the EOS mission “Aura spacecraft” since July 2004. OMI is a high spatial resolution (13km×24km) ground pixel size ultraviolet/visible (UV/VIS) backscatter spectrometer (Levelt et al., 2006). The OMI aerosol retrieval algorithm is the same as for the Total Ozone Mapping spectrometer (TOMS) near-UV method of aerosol absorption sensing from space (Torres et al., 2005). The accuracy of the OMI retrieval of aerosol optical depth is around 30 % relative to AERONET measurements (Torres et al., 2005).

The extinction AOD at 500 nm (near-UV) has been selected from the OMI product from Giovanni site portal (http://Giovanni.gsfc.nasa.gov/giovanni). From that data we selected 2∘×2∘ box around the three AERONET stations (see Fig. 1) for comparison purposes with the observations. The time span of this data is from January 2008 to December 2011.

The International Centre for Theoretical Physics (ICTP) Regional Climate Model (RegCM) was built upon the National Center for Atmospheric Research (NCAR) Mesoscale Model version 4 (MM4) (Giorgi and Bates, 1989; Giorgi et al., 1993a, b).

The Regional Climate Model version 4 (RegCM4) is the second major development of the RegCM core after RegCM3 (Pal et al., 2007). The coding structure is completely changed. It has become totally FORTRAN 90 compliant and modular structured and its parallelization and memory management has become more efficient. The RegCM4 has more physics options, which include the Community Land surface Model (CLM3.5) land surface parameterization, the Tiedtke convection scheme, the University of Washington (UW) planetary boundary layer (PBL) scheme, and the Rapid Radiative Transfer Model (RRTM) (Giorgi et al., 2012).

RegCM4 is an online climate-chemistry model and has an online gas-phase chemistry scheme (CBMZ) (Shalaby et al., 2012). It has various aerosol components such as, four size bin dust, two size bin sea salt, sulphate, black carbon and organic carbon. Sulphate, black carbon and organic carbon (Solmon et al., 2006), as well as dust and sea-salt are radiatively active (Zakey et al., 2006, 2008).

Sand particles are affected by many forces that determine their fate. Dust particles have three dynamic modes: (a) saltation: small particles move by jumping like leap-frog, Once lifted by wind it will drift downwind and return to hit the ground again and transfer energy and momentum to other soil aggregates (soil particles). (b) Creeping: large dust particles cannot be lifted into the air, but will just move and slide on the ground. (c) Suspension: if the upward draft is strong enough to compensate the gravitational force of the dust particles, dust will remain airborne and be transported by the wind over longer distances until its gravitational force overcome the uplifting force. It is believed that saltation is the main mechanism for surface dust emission (Shao et al., 1993; Marticorena and Bergametti, 1995).

Following Marticorena and Bergametti (1995) and Alfaro and Gomes (2001), a complex dust emission scheme has been implemented in RegCM3 (Zakey et al., 2006). This emission scheme is based on parameterization of soil aggregate saltation and sandblasting processes. According to this scheme, a critical parameter for the dust saltation process is the threshold friction velocity ut∗, which is a function of particle size Dp (Eq. 1), such that, uts∗ represents an ideal minimum threshold friction velocity, feff is a correction factor accounting for the effect of surface roughness and fw is a factor that accounts for the effect of soil moisture content on the threshold friction velocity. The particle size is determined by the land surface soil texture. Calculating the threshold friction velocity is required to calculate the horizontal dust flux (dHF) (Eq. 2), such that, E is the ratio of erodible to total surface, dSrel is the relative surface of soil aggregate of diameter Dp to the total aggregate surface and R(Dp) is the ratio of the threshold friction velocity defined in (Eq. 1) to the friction velocity u∗ calculated within each grid cell from the model prognostic surface wind and the surface roughness. Finally, the vertical dust emission flux dFkin is assumed to be directly proportional to the horizontal flux dHF (Eq. 3), such that, β is an empirical factor and its value is 16 300 cms-2 (Zakey et al., 2006). ut∗Dp=uts∗DpfefffwdHFDp=Eρagu∗31+RDp(1-R2Dp)dSrelDpdFkinRp=βdHFDp Where, g is the gravity acceleration constant and ρa is air density. RegCM3 (a previous version) has been used frequently to simulate atmospheric dust and its radiative impact. Most of these simulations have been performed for areas in the Sahara (Zakey et al., 2006; Konarè et al., 2008; Solmon et al., 2008, 2012; Napat et al., 2012; Steiner et al., 2014), apart from one over East Asia (Zhang et al., 2009). Over the Middle East and Arabian Peninsula there have been a few studies such as (Marcella and Eltahir, 2010, 2011, 2012; Nazrul Islam and Almazroui, 2012).

The model simulations are conducted for 14 years (1999–2012), but the analysis is done for the 2000–2012 time period with the first year used for model spin-up. The European Centre for Medium Range Weather Forecasts Reanalysis project (ERA-Interim) reanalysis data (https://apps.ecmwf.int/datasets/data/interim_full_daily) has been used to provide the model with 6 h meteorological boundary conditions (Dee et al., 2011). Monthly climatology of dust aerosol is provided by the Model for Ozone and Related chemical Tracer (MOZART) model at 2.8∘×2.8∘ (T42) resolution (Emmons et al., 2010).

Figure 1 shows the model domain at 50 km grid resolution. The model vertical resolution is 18 sigma levels. The model physics is configured as follows; the boundary layer scheme is UW-PBL, this is a 1.5 order local, down-gradient diffusion parameterization in which the velocity scale is based on the turbulent kinetic energy (TKE). The TKE in turn is calculated prognostically from the balance of buoyant production/destruction, shear production, dissipation vertical transport, and horizontal diffusion and advection (Giorgi et al., 2012; O'Brien et al., 2012).

The Bio-sphere-Atmosphere Transfer scheme (BATS) (Dickinson et al., 1993) is used to parameterize the land-surface atmosphere interaction. The full scheme includes a 1-layer vegetation module, a 1-layer snow module, a force restore model for soil temperature, a 3-layer soil scheme and a simple surface runoff parameterization. This scheme is simpler than the other CLM3.5 schemes in the model, which have many physical parameterizations of soil processes (Tawfik and Steiner, 2011). It does not require time for the soil to equilibrate with the atmosphere (Giorgi et al., 2012). The convection scheme provided by Grell (Grell, 1993) is used along with the Community Climate Model version 3 (CCM3) radiation scheme (Kiehl et al., 1996).

Figure 1a and b shows the BATS land texture and topography of the region. BATS land texture constitute of 17 categories (see Table 2 in Zakey et al., 2006). The first 16 categories represent the soil texture based on the classical sand-silt-clay percentage approach (Hillel, 2003). The stars indicate the location of the AERONET stations that will be used for model validation. The area around the AERONET stations will be used to calculate the area average of the modeled AOD for comparison against observations. S1 represents the Mesopotamian source region that includes Iraq and Arabian Peninsula, S2 represents the Red Sea source region that includes Al nofod desert, S3 represents the El Rob El khali desert, S4 represents the Somalia desert source region and S5 represents the southwest Asia source region that includes dry beds in the Afghanistan-Pakistan-Iran border.

The Kuwait University site is near major dust sources (S1 and S2 in Fig. 1a) and would be representative to the northern Arabian Peninsula dust aerosol climatology. Figure 4 shows a 5 year statistics of AOD for AERONET station, a 7 years statistics of AOD for the satellite observations and a 13 year statistics of AOD for RegCM4.

Table 1 shows that the AERONET annual averaged median is ∼0.4. Therefore the dust season months, which exceed 20 % of the annual averaged median are April–July (Table 1). May shows the maximum AOD, which is 50 % higher than the annual average. The May–July months show a maximum variability (the difference between 25 percentile and 75 percentile is the greatest). December and January show a minimum AOD of 0.2 (Fig. 4).

The observational dataset shows some disagreement about the length and intensity of the dust season according to the proposed criterion. AERONET and MISR show four months as dust season, while MODIS and OMI only show three month as dust season. In addition AERONET and MISR AOD values are often higher than the MODIS and OMI AOD values.

The Solar-Village site would be a representative of central Arabian Peninsula dust aerosol climatology. The altitude of this site (764 m a.s.l) and its location in the vicinity of complex terrain may contribute to the apparent differences in AOD climatology with regard to the other sites. Figure 5 shows a 13 year statistics of AERONET AOD, a 7 year statistics of AOD for the satellite observations and a 13 year statistics of RegCM4 AOD.

The satellite group shows maximum AOD in March–June, except for OMI, whose maximum AOD occur in February–August. In addition, OMI has higher AOD values than the other satellite platforms. AERONET shows maximum median in April–August months and underestimates the AOD (Fig. 5).

The AERONET AOD statistics show that the AOD annual average median is 0.27 (Table 1). Since dust concentration decreases with height, therefore The AERONET AOD of Solar-Village is much less than the Kuwait University site and this may be due to the high altitude of Solar-Village and its far distance to major dust sources. Nevertheless, this station could capture a severe dust episode as reported by Alharbi et al. (2013) and Kalenderski et al. (2013).

The dust season in this region spans five months (April–August) according to the AERONET data; however, MODIS, MISR and OMI show April–June months as the dust season (Table 3). MODIS, MISR and OMI show higher AOD than AERONET for Solar-Village at the time of the dust season (Fig. 5).

Solar-Village shows a different climatology in terms of the dust season onset and its intensity. The model shows a larger temporal extension of the dust season and an early onset; it starts in March and lasts until August.

The Mezaira site is near the El rob El khali desert, which is a major dust source in the Arabian Peninsula (S3 in Fig. 1a). The site is representative for southern Arabian Peninsula dust aerosol climatology.

Figure 6 shows a 7 year statistics of AERONET AOD, 7 year statistics of AOD for the satellite observations and a13 year statistics of AOD for RegCM4.

The annual average median is ∼0.4 (Table 1). The extension of the dust season varies among the datasets, AERONET shows 5 months dust season (April–August), MISR and OMI show 4 months dust season (May–August), MODIS shows 3 months dust season (June–August) and finally RegCM4 shows a 2 months dust season (July–August). They all agree about the end of the dust season in August but they disagree on the dust season onset (Table 1).

Different regions in the Arabian Peninsula have different dust aerosol climatology. The main feature of this climatology is the forward time shift of the dust season towards the south of the Arabian Peninsula. This feature is evident in Fig. 3. The quantitative analysis in Table 1 also reflects this phenomenon in all datasets. Table 1 shows how the dust season is shifted towards summer months along the traverse from the northern (Kuwait site) to the southern part of the Arabian Peninsula (UAE site). This feature will be explained by the analysis of the atmospheric dynamics in the following section.

The validation of the modeled AOD against the observations shows how the model behaves differently from region to region and could shed some light on the sources of uncertainty of the model. Table 2 (first column for each site) lists the modeled AOD monthly medians for the three sites and it is used to calculate the relative error with respect to each observational datasets according to Eq. (5).

RegCM4 underestimates AOD over the northern part of the Arabian Peninsula with respect to all observational datasets (Table 2). For the dust season months (April–July) the model relative error ranges between 29 and 68 %. This model feature is also noticed in Marcella and Eltahir (2010). Toward the central Arabian Peninsula, the model performance becomes better, and the relative error is between 7 and 60 %. The model underestimates in the late spring season, but tends to overestimate in the summer season (Table 2). Toward the southern Arabian Peninsula, the model error ranges reduced compared to the other stations, for instance, the model errors in June are 1 and 5 % with respect to AERONET and MODIS respectively. On the other hand, the model shows significant over estimation in August, which is as high as 90 and 75 % with respect to AERONET and MODIS, respectively (Table 2).

In winter and early spring (January–April) the Arabian Peninsula is affected by an extension of Siberian High pressure and the Red Sea trough in northern part and southern part of the Arabian Peninsula respectively. The anti-cyclonic wind field is dominated over the Arabian Peninsula. Figure 7a and b shows the climatology of the mean sea level pressure (MSLP), while Fig. 7d and e shows the climatology of temperature and wind field at 850 hPa geopotential surface. While the extension of Siberian high pressure with cold air mass affects the northern part of the Arabian Peninsula (Fig. 7a and b), the southerly winds which are associated with the Red Sea trough over the southern Arabian Peninsula advect warm air toward north (Fig. 7d and e), which leads to instabilities in the atmosphere.

The position and strength of the Sub-Tropical Jet (STJ) at 200 hPa determine the atmospheric stability at the surface. Figure 7g and h shows the climatology of the STJ. In winter the STJ core is located over the Arabian Peninsula. The spring STJ core is weaker than the winter STJ. It migrates toward north in April–May and displays a strong meridional component (Fig. 7g and h). Alhabri et al. (2013) described the role of the STJ in the onset and development of dust storms. Briefly, the STJ generates regions of upward motion and downward motion (i.e. secondary circulation). The north side of the STJ core is a divergence zone that is associated with upward motion, while the south of the STJ core is a conversion zone that is associated with downward motion. Figure 8 shows the zonally averaged vertical velocity (Omega in hPas-1 where (-) values designate upward motion and (+) values designate downward motion). In December–April the STJ core is located south of 30∘ N (Fig. 7g and h), and Fig. 8 shows the secondary circulation region of upward motion and downward motion from surface (1000 hPa) to the upper air (100 hPa) south of 30∘ N. In late winter (February) and spring time (March–April) the upward motion progressively strengthens from 20∘ N to 30∘ N and from surface to upper air levels.

In most part of the Arabian Peninsula, the prevailing wind directions are southerlies and southwesterlies. The observations over the northern and southern parts of the Arabian Peninsula show that during the dust season, the prevailing wind directions are southerlies and southwesterlies (Mashat et al., 2008). Figure 9 shows the climatology of the zonally averaged wind's meridional component. The southerly meridional component shows the ascending motion and the sliding of warm air over the cold air from the north.

Figure 10 shows the climatology of the zonally averaged vertical profile of fine dust concentration. The maximum concentration is located between 15 and 30∘ N. The dust concentrations progressively increase towards north from December to February and start to retreat in April. In winter and spring, dust layers do not extend to high levels in the atmosphere, actually they are confined to levels below 800 hPa.

The friction velocity is the sole dynamical variable in the dust parameterization (the other variables are land surface characteristics). The land surface response of the large scale circulation is reflected by the friction velocity. Also, according to Eqs. (2) and (3), the dust emission is proportional to the friction velocity. In Supplement Fig. S1 the friction velocity over land surfaces is shown and in the following discussion, we will focus in particular on the dust source regions (Fig. 1a).

In the late winter and early spring months (January–March) the friction velocity is high in the western and central part of the Arabian Peninsula (Fig. S1). Correspondingly, dust emissions are also higher in the western and central Arabian Peninsula compared to the surrounding areas (Fig. 11). Over the source region S4 (Al Roub AL khali desert) dust emission is clearly accompanied with high friction velocity. In the northern part of the Arabian Peninsula the climatology of the friction velocity does not reflect the abovementioned relation between friction velocity and surface dust emission. Although, this region is affected by a high pressure system which causes a weak surface wind field, most of the emitted dust in this region is due to a fast migrating cyclone over the Syria–Iraq area associated with high surface wind (Mashat et al., 2008; Abdi Vishkaee et al., 2012; Alharbi et al., 2013) that has been filtered out in the averaging procedure.

The corresponding atmospheric surface dust concentration for dust less than 10 µm is shown in Fig. S2. In the months from December to April the surface dust concentration over the Arabian Peninsula reaches up to 120 µgm-3 in the central region, which is in agreement with results for the same region performed in a global modeling study by Ginoux et al. (2004).

Starting mid-spring (i.e. April) the surface high pressure retreats and the Indian Monsoon depression starts to progress towards the Arabian Peninsula. In summer (June–August) the Arabian Peninsula is largely impacted by the Indian Monsoon depression and it further intensifies during summer months (Fig. 7c). The prevailing winds in this season are northerlies and northwesterlies (“Shamal”) over the most of the Arabian peninsula (Mashat et al., 2008), while the southern part is affected by strong southerly winds associated with the Low-Level Jet (i.e. Somalia Jet) that is triggered by the Indian Monsoon (Fig. 7f). In late spring and summer (i.e. April–August), the STJ migrates toward north and a high pressure is develops over the Arabian Peninsula (Fig. 7i).

The summer meridional components in Fig. 9 become northerly, which is an indication of the Indian Monsoon cyclonic circulation. In addition, Fig. 8 shows that the ascending motion in the southern part of the Arabain Peninsula (15–25∘ N) is higher than during the winter-spring time. The descending motion over the ocean is evident in the summer season at latitudes from 10 to 15∘ N. According to Fig. 10 the ascending motion in southern Arabia in upper air results in uplifting of dust to higher atmospheric levels up to the 500 hPa level. After the retreat of the Indian Monsoon, the ascending motion weakens significantly and the meridional velocity reverses its sign to be southerly again.

During the summer season, starting June, the friction velocity (Fig. S1) gets higher in the eastern and southern part of the Arabian Peninsula, and over Somalia (dust sources S3 and S4 in Fig. 1a). As a consequence high surface dust emission fluxes (Fig. 11) occur over the eastern and the southern Arabian Peninsula (e.g. in Oman) and over Somalia. Those source regions contribute significantly to the resultant AOD in summer. The surface dust concentration in summer reaches maximum values between 100–180 µgm-3 (Fig. S2).